**Part 2**

**Diagnostic Markers** 

90 Prostate Cancer – Original Scientific Reports and Case Studies

Park, S.Y., Kim, Y.J., Gao, A.C., Mohler, J.L., Onate, S.A., Hidalgo, A.A., Ip, C., Park, E.M.,

Reiter, R.J. (1998) Cytoprotective properties of melatonin: presumed association with

Reiter, R.J., Tan, D.X., Manchester, L.C.& Qi, W. (2001) Biochemical reactivity of melatonin

Semenza, G.L.& Prabhakar, N.R. (2007) HIF-1-dependent respiratory, cardiovascular, and

Shiu, S.Y., Li, L., Xu, J.N., Pang, C.S., Wong, J.T.& Pang, S.F. (1999) Melatonin-induced

Siu, S.W., Lau, K.W., Tam, P.C.& Shiu, S.Y. (2002) Melatonin and prostate cancer cell

Sreekumar, A., Poisson, L.M., Rajendiran, T.M., Khan, A.P., Cao, Q., Yu, J., Laxman, B.,

Tamarkin, L., Baird, C.J.& Almeida, O.F. (1985) Melatonin: a coordinating signal for

Tan, D.X., Manchester, L.C., Reiter, R.J., Qi, W.B., Karbownik, M.& Calvo, J.R. (2000)

Terraneo, L., Bianciardi, P., Caretti, A., Ronchi, R.& Samaja, M. (2010) Chronic systemic

Vaupel, P. (2004) Tumor microenvironmental physiology and its implications for radiation

Xi, S.C., Siu, S.W., Fong, S.W.& Shiu, S.Y. (2001) Inhibition of androgen-sensitive LNCaP

Zhong, H., De Marzo, A., Laughner, E., Lim, M., Hilson, D.& Zagzag, D. (1999)

Zhong, H., Semenza, G.L., Simons, J.W.& De Marzo, A.M. (2004) Up-regulation of hypoxia-

prostate cancer cells. *Cancer Res* 66(10)**:** 5121-5129.

*Biophys* 34(2)**:** 237-256.

*Res* 27(3)**:** 183-192.

sensitivity. *Prostate* 52(2)**:** 106-122.

progression. *Nature* 457(7231)**:** 910-914.

oncology. *Semin Radiat Oncol* 14(3)**:** 198-206.

cancer cells. *J Pineal Res* 29(3)**:** 172-183.

*Prev* 28(2)**:** 88-93.

their metastases. *Cancer Res* 59**:** 5830-5835.

*Biol Signals Recept* 9(3-4)**:** 137-159.

1254.

mammalian reproduction? *Science* 227(4688)**:** 714-720.

Wilding, G. (1995) Endocrine control of prostate cancer. *Cancer Surv* 23**:** 43-62.

oxidative damage and aging. *Nutrition* 14(9)**:** 691-696.

Yoon, S.Y.& Park, Y.M. (2006) Hypoxia increases androgen receptor activity in

with reactive oxygen and nitrogen species: a review of the evidence. *Cell Biochem* 

redox responses to chronic intermittent hypoxia. *Antioxid Redox Signal* 9(9)**:** 1391-1396.

inhibition of proliferation and G1/S cell cycle transition delay of human choriocarcinoma JAr cells: possible involvement of MT2 (MEL1B) receptor. *J Pineal* 

proliferation: interplay with castration, epidermal growth factor, and androgen

Mehra, R., Lonigro, R.J., Li, Y., Nyati, M.K., Ahsan, A., Kalyana-Sundaram, S., Han, B., Cao, X., Byun, J., Omenn, G.S., Ghosh, D., Pennathur, S., Alexander, D.C., Berger, A., Shuster, J.R., Wei, J.T., Varambally, S., Beecher, C.& Chinnaiyan, A.M. (2009) Metabolomic profiles delineate potential role for sarcosine in prostate cancer

Significance of melatonin in antioxidative defense system: reactions and products.

hypoxia promotes LNCaP prostate cancer growth in vivo. *Prostate* 70(11)**:** 1243-

prostate cancer growth in vivo by melatonin: association of antiproliferative action of the pineal hormone with mt1 receptor protein expression. *Prostate* 46(1)**:** 52-61. Xi, S.C., Tam, P.C., Brown, G.M., Pang, S.F.& Shiu, S.Y. (2000) Potential involvement of mt1

receptor and attenuated sex steroid-induced calcium influx in the direct antiproliferative action of melatonin on androgen-responsive LNCaP human prostate

Overexpression of hypoxia inducible factor 1alpha in common human cancers and

inducible factor 1alpha is an early event in prostate carcinogenesis. *Cancer Detect* 

**6** 

*Malaysia* 

**Cancer Detection from Transrectal Ultrasound** 

Selvalingam S., Leong A.C., Natarajan C., Yunus R.1 and Sundram M.

Transrectal ultrasound-guided biopsy of the prostate is the mainstay in the diagnosis of prostate cancer. Cancer detection rates varies from centre to centre and is dependent on various factors including technique, number of cores, prostate volume, PSA levels and

There are also differences in various ethnic groups with regards to prostate cancer incidence especially in the west where African Americans have a higher incidence. In Malaysia there are three major ethnic groups; Malays (65%), Chinese(25%) and Indians (8%). There is no evidence as yet to show any differences in prostate cancer detection

The Malaysian National Cancer Registry published in 2006¹, ranks prostate cancer as the 4th most common cancer in Malaysian men after large bowel, lung and nasopharyngeal cancer. It constitutes 7.3% of all cancers in men. The overall prostate cancer incidence per 100 000 population (CR) was 7.3 and Age Standardised Incidence (ASR) was 12. By ethnicity Malays had the lowest ASR , 7.7 followed by Indians, 14.8 and Chinese 15.8. In fact Prostate Cancer is the fifth most common cancer among Malay men, fourth among Chinese and the second most common cancer among the Malaysian Indians. Overall , the Age Specific Incidence per 100 000 population increased from 9.7 among men in their 50s

In an observation by Lim et al¹ from the Malaysian Clinical Research Center, Malaysian Indians have a higher incidence compared to Indians from Chennai, Malaysian Malays have a lower incidence compared to Singapore Malays and Malaysian Chinese have the highest incidence compared to Chinese in other Asian countries. These findings will need to be

It is also postulated that Asians prostate are generally smaller than western counterparts but volume for volume Asians have a higher PSA and part of the reason may be due to higher

The primary objective of this study is to look at a single centre's cancer detection rate and to determine the various factors that may influence cancer detection namely age, PSA levels,

**1. Introduction** 

digital rectal examination findings.

among the three ethnic groups

to 60.4 in men in their 60s

verified by further research.

**2. Aims and objectives** 

level of inflammation in Asian prostates.

**Guided Biopsy in a Single Center** 

*Department of Urology, General Hospital Kuala Lumpur 1Department of Pathology, General Hospital Kuala Lumpur* 

### **Cancer Detection from Transrectal Ultrasound Guided Biopsy in a Single Center**

Selvalingam S., Leong A.C., Natarajan C., Yunus R.1 and Sundram M. *Department of Urology, General Hospital Kuala Lumpur 1Department of Pathology, General Hospital Kuala Lumpur Malaysia* 

#### **1. Introduction**

Transrectal ultrasound-guided biopsy of the prostate is the mainstay in the diagnosis of prostate cancer. Cancer detection rates varies from centre to centre and is dependent on various factors including technique, number of cores, prostate volume, PSA levels and digital rectal examination findings.

There are also differences in various ethnic groups with regards to prostate cancer incidence especially in the west where African Americans have a higher incidence. In Malaysia there are three major ethnic groups; Malays (65%), Chinese(25%) and Indians (8%). There is no evidence as yet to show any differences in prostate cancer detection among the three ethnic groups

The Malaysian National Cancer Registry published in 2006¹, ranks prostate cancer as the 4th most common cancer in Malaysian men after large bowel, lung and nasopharyngeal cancer. It constitutes 7.3% of all cancers in men. The overall prostate cancer incidence per 100 000 population (CR) was 7.3 and Age Standardised Incidence (ASR) was 12. By ethnicity Malays had the lowest ASR , 7.7 followed by Indians, 14.8 and Chinese 15.8. In fact Prostate Cancer is the fifth most common cancer among Malay men, fourth among Chinese and the second most common cancer among the Malaysian Indians. Overall , the Age Specific Incidence per 100 000 population increased from 9.7 among men in their 50s to 60.4 in men in their 60s

In an observation by Lim et al¹ from the Malaysian Clinical Research Center, Malaysian Indians have a higher incidence compared to Indians from Chennai, Malaysian Malays have a lower incidence compared to Singapore Malays and Malaysian Chinese have the highest incidence compared to Chinese in other Asian countries. These findings will need to be verified by further research.

It is also postulated that Asians prostate are generally smaller than western counterparts but volume for volume Asians have a higher PSA and part of the reason may be due to higher level of inflammation in Asian prostates.

#### **2. Aims and objectives**

The primary objective of this study is to look at a single centre's cancer detection rate and to determine the various factors that may influence cancer detection namely age, PSA levels,

Cancer Detection from Transrectal Ultrasound Guided Biopsy in a Single Center 95

Age Stratification

Fig. 1. Stratification of age groups among patients who were biopsied.

Table 1. Increasing Prostate Volume seen with Increasing Age

Age stratification (% Patients)

<50 50-60 61-70 71-80

<20 0% 17.6 % 4.2 % 2.3 %

20-29 0% 23.5% 8.5% 15.9 %

30-39 100% 5.9% 21.1% 18.2%

40-49 0% 23.5% 25.4 % 15.9%

>50 0% 29.4% 40.8% 47.7%

The median PSA of patients younger than 50 years old was 6.8 +/- 4.7 ng/mL , patients 51 to 60 years, 7.96 +/- 177 ng/mL, patients 61 to 70 years , 7.89 +/- 124 ng/mL, patients 71 to 80 years old, 11.0 +/- 127 ng/mL and patients older than 80 years, 26.5 +/- 163ng/mL. (Figure 2) With regards to race, the Malay patients had a median PSA 10 +/- 138.4ng/mL , Chinese patients median PSA 8.55 +/- 113.8 ng/mL and Indians had the lowest median PSA

Prostate volume (g)

7.6 +/- 167 ng/mL.

number of cores and prostate volume. The study also aims to determine if cancer detection is higher in Chinese and Indian patients compared to Malay patients as reflected by the national Age Standardised Incidence in the three ethnic groups.

#### **2.1 The secondary objectives include**

To determine the detection of prostate inflammation and PIN and its correlation with age, PSA levels , prostate volume and ethnic group within this small cohort of patients.

To evaluate if prostates with malignancy have a strong association with inflammation and PIN.

To determine if malignancy in older patients is more aggressive as reflected by a higher Gleason sum and higher PSA.

### **3. Materials and methods**

671 patients who underwent TRUS biopsies of the prostate from January 2009 to August 2010 were analyzed and the various parameters associated with each biopsy documented. These included patient demographics such as age, race, PSA and previous biopsy history. Prostate parameters included size, digital rectal examinations findings and number of cores taken. The histological parameters looked at were Gleason primary and secondary scores and total percentage of tumour.

Transrectal Ultrasound Guided biopsies were performed by various operators ranging from urological trainees to consultants. There was variablility in the number of cores taken where a few operators were following the Vienna nomogram and others were doing a standard 12 core biopsy. Patients who had more than 12 cores were either having a repeat biopsy or had additional targeted biopsies based on ultrasound findings. 95.8% of patients evaluated were undergoing their first biopsy. Prostate volume was assessed transrectally using the BK Hawk Ultrasound. Prostate volume was available for analysis only from January 2010 onwards.

Statistical analysis was with SPSS version 18, Chi square test was used for categorical data and independent t test used to compare means.

#### **4. Results**

Between January 2009 to August 2010, a total of 671 TRUS biopsy results were analysed. The mean age of patients presenting for TRUS biopsy at our centre was 68.38 +/- 7years. Overall median PSA was 9 +/- 132.9 ng/mL .The ethnic distribution of patients included 48.1% Malays, 36.7% Chinese, 13.1% Indians and 1.6% of other ethnic origin. Compared to the national demographics there were less Malays and more Chinese and Indians in this cohort of patients. 50.5% of our patients presented with a prostate specific antigen (PSA) level of between 4 to 10ng/mL and 24.7% presented with levels higher than 20 ng/mL.

The majority of patients had a reasonably high prostate volume; 41.6% had a volume of more than 50g while only 17.5% had a prostate volume less than 30g. Overall Malay and Indian patients presented with larger prostates. 90% of Malay and Indian patients presented with prostate volume of more than 30g compared to 66.7% of Chinese patients.

There is an increasing trend of prostate volume and PSA level with age. None of the patients less than 50 years old had a prostate volume of more than 50g, while 48% of patients older than 70 years old had volume more than 50g. (Table 1)

number of cores and prostate volume. The study also aims to determine if cancer detection is higher in Chinese and Indian patients compared to Malay patients as reflected by the

To determine the detection of prostate inflammation and PIN and its correlation with age,

To evaluate if prostates with malignancy have a strong association with inflammation and

To determine if malignancy in older patients is more aggressive as reflected by a higher

671 patients who underwent TRUS biopsies of the prostate from January 2009 to August 2010 were analyzed and the various parameters associated with each biopsy documented. These included patient demographics such as age, race, PSA and previous biopsy history. Prostate parameters included size, digital rectal examinations findings and number of cores taken. The histological parameters looked at were Gleason primary and secondary scores

Transrectal Ultrasound Guided biopsies were performed by various operators ranging from urological trainees to consultants. There was variablility in the number of cores taken where a few operators were following the Vienna nomogram and others were doing a standard 12 core biopsy. Patients who had more than 12 cores were either having a repeat biopsy or had additional targeted biopsies based on ultrasound findings. 95.8% of patients evaluated were undergoing their first biopsy. Prostate volume was assessed transrectally using the BK Hawk Ultrasound. Prostate volume was available for analysis only from January 2010

Statistical analysis was with SPSS version 18, Chi square test was used for categorical data

Between January 2009 to August 2010, a total of 671 TRUS biopsy results were analysed. The mean age of patients presenting for TRUS biopsy at our centre was 68.38 +/- 7years. Overall median PSA was 9 +/- 132.9 ng/mL .The ethnic distribution of patients included 48.1% Malays, 36.7% Chinese, 13.1% Indians and 1.6% of other ethnic origin. Compared to the national demographics there were less Malays and more Chinese and Indians in this cohort of patients. 50.5% of our patients presented with a prostate specific antigen (PSA) level of

The majority of patients had a reasonably high prostate volume; 41.6% had a volume of more than 50g while only 17.5% had a prostate volume less than 30g. Overall Malay and Indian patients presented with larger prostates. 90% of Malay and Indian patients presented

There is an increasing trend of prostate volume and PSA level with age. None of the patients less than 50 years old had a prostate volume of more than 50g, while 48% of patients older

between 4 to 10ng/mL and 24.7% presented with levels higher than 20 ng/mL.

with prostate volume of more than 30g compared to 66.7% of Chinese patients.

PSA levels , prostate volume and ethnic group within this small cohort of patients.

national Age Standardised Incidence in the three ethnic groups.

**2.1 The secondary objectives include** 

Gleason sum and higher PSA.

**3. Materials and methods** 

and total percentage of tumour.

and independent t test used to compare means.

than 70 years old had volume more than 50g. (Table 1)

PIN.

onwards.

**4. Results** 

Fig. 1. Stratification of age groups among patients who were biopsied.


Table 1. Increasing Prostate Volume seen with Increasing Age

The median PSA of patients younger than 50 years old was 6.8 +/- 4.7 ng/mL , patients 51 to 60 years, 7.96 +/- 177 ng/mL, patients 61 to 70 years , 7.89 +/- 124 ng/mL, patients 71 to 80 years old, 11.0 +/- 127 ng/mL and patients older than 80 years, 26.5 +/- 163ng/mL. (Figure 2) With regards to race, the Malay patients had a median PSA 10 +/- 138.4ng/mL , Chinese patients median PSA 8.55 +/- 113.8 ng/mL and Indians had the lowest median PSA 7.6 +/- 167 ng/mL.

Cancer Detection from Transrectal Ultrasound Guided Biopsy in a Single Center 97

Histology

Fig. 3. Variability of biopsy histology based on age stratification. Increasing detection of

With regards to PSA at presentation, malignancy was detected in 15.3% of patients with PSA 0 to 10ng/ml, 16.2% with PSA 11 to 20ng/ml, 39% with PSA 21 to 50ng/ml and 77.9% with PSA >50ng/ml. Inflammation was seen in 13.8% of patients with PSA 0 to 10ng/ml, 27% with PSA 11 to 20ng/ml,24.7% with PSA 21 to 50ng/ml and 6.5% with PSA more than 50ng/ml. PIN was seen in 11% of patients with PSA 0 to 10ng/ml, 12.6% with PSA 11 to 20ng/ml, 3.9% with PSA 21 to 50ng/ml and 0% with PSA more than 50ng/ml

As expected, patients diagnosed with malignancy had a higher median PSA of 26.6 +/- 230ng/ml as compared with benign disease (7.3 +/- 82ng/mL,) inflammation (11.7 +/- 23.7ng/mL) and PIN (8.2 +/- 7.1ng/mL). In our study, the PSA level did not show any

The higher the prostate volume, the lower the cancer detection. 60% of prostates weighing less than 20g were found to be malignant, 35.3% malignant for volume 20 to 29g, 24% for volume 30 to 39g, 21.4% for volume 40 to 49g and 14.5% for volume more than 50g. PIN was generally not seen in prostates less than 40g. Inflammation however, was seen in 14 to 17%

Of the 25.6% of patients with malignancy, 28% of them had a Gleason sum of 6, 24.8% with Gleason 7, 22.4% with Gleason 8, 24.2% with Gleason 9 and 0.6% with Gleason 10. Gleason sum 7 was seen in 30.8%, 19.3% and 28.6% of patients in their 6th, 7th and 8th decade of life respectively. Gleason sum 8 and above was seen in 60.6%, 43.9% and 43.1% of patients in their 6th,7th and 8th decade of life respectively. Patients above 80 years presented with

significantly higher grade disease. 83.3% had gleason sum 8 and above.

malignancy with increasing age.

correlation with prostate volume.

of cases regardless of the prostate volume.

(Figure 4)

Fig. 2. Increasing PSA with increasing Median Age among the patients biopsied

In a separate cross sectional study done among Malaysian community in 2005², as part of a prostate awareness campaign, the mean PSA of men in their 50s was 1.4 +/- 6.3 ng/mL , men in their 60s, 2.3 +/- 3.8 ng/mLand above 70, 4.3 +/- 11 ng/mL. In the similar study Malays had a mean PSA 2.3 +/- 8.3 ng/mL, Chinese 1.8 +/- 3 ng/mL and Indians 1.3 +/- 1.9 ng/mL.

This ethnic variation in PSA in Malaysia is contrary to a separate study by Chia et al 3 from Singapore who did not find any PSA variation between the three ethnic groups, however in that study 92.8% of participants were Chinese.

From the 671 biopsies analysed, 29.1% had 6 to 10 cores taken, 46.8% had 12 cores, and 24.1% had more than 12 cores. The TRUS biopsy cancer detection rate at our center was 25.6% and it was almost similar in all the major ethnic groups (Malay-24%, Chinese- 26.2%, Indian 24.4%)(p> 0.05). Prostate inflammation was identified in 16.8% of our patients, while PIN was seen in 9%. Prostate inflammation was fairly similar in the Malay and Chinese population (Malay-17.4%, Chinese-17.8%) but the Indian population had a lower inflammation rate of 12.8%.

Of the patients who had malignancy, 34.2% had 6 to 10 cores taken, 46.2% had 12 cores and 19.6% had more than 12 cores taken. When compared with patients who were diagnosed with benign disease, the distribution of cores taken was similar;26.5% with 6 to 10 cores taken, 51.6% with 12 cores and 21.9% with more than 12 cores.

Cancer detection was 0 % in patients < 50 years, 17.6% in those 51 to 60 years old, 20.2% in patients 61 to 70 years old, 32.4% for those 71 to 80 years old and 50% for patients older than 80 years. There was no PIN seen in patients < 50 years old, 4.1% in patients 51 to 60 years old, 11.1 % for those 61 to 70 years old, and 8.3% above the age of 70 years. Prostate inflammation was seen in all age groups ranging from 11% to 19% and lowest (8.3%) in patients older than 80 years (Figure 3)

Fig. 2. Increasing PSA with increasing Median Age among the patients biopsied

that study 92.8% of participants were Chinese.

taken, 51.6% with 12 cores and 21.9% with more than 12 cores.

inflammation rate of 12.8%.

patients older than 80 years (Figure 3)

In a separate cross sectional study done among Malaysian community in 2005², as part of a prostate awareness campaign, the mean PSA of men in their 50s was 1.4 +/- 6.3 ng/mL , men in their 60s, 2.3 +/- 3.8 ng/mLand above 70, 4.3 +/- 11 ng/mL. In the similar study Malays had a mean PSA 2.3 +/- 8.3 ng/mL, Chinese 1.8 +/- 3 ng/mL and Indians 1.3 +/- 1.9 ng/mL. This ethnic variation in PSA in Malaysia is contrary to a separate study by Chia et al 3 from Singapore who did not find any PSA variation between the three ethnic groups, however in

From the 671 biopsies analysed, 29.1% had 6 to 10 cores taken, 46.8% had 12 cores, and 24.1% had more than 12 cores. The TRUS biopsy cancer detection rate at our center was 25.6% and it was almost similar in all the major ethnic groups (Malay-24%, Chinese- 26.2%, Indian 24.4%)(p> 0.05). Prostate inflammation was identified in 16.8% of our patients, while PIN was seen in 9%. Prostate inflammation was fairly similar in the Malay and Chinese population (Malay-17.4%, Chinese-17.8%) but the Indian population had a lower

Of the patients who had malignancy, 34.2% had 6 to 10 cores taken, 46.2% had 12 cores and 19.6% had more than 12 cores taken. When compared with patients who were diagnosed with benign disease, the distribution of cores taken was similar;26.5% with 6 to 10 cores

Cancer detection was 0 % in patients < 50 years, 17.6% in those 51 to 60 years old, 20.2% in patients 61 to 70 years old, 32.4% for those 71 to 80 years old and 50% for patients older than 80 years. There was no PIN seen in patients < 50 years old, 4.1% in patients 51 to 60 years old, 11.1 % for those 61 to 70 years old, and 8.3% above the age of 70 years. Prostate inflammation was seen in all age groups ranging from 11% to 19% and lowest (8.3%) in Fig. 3. Variability of biopsy histology based on age stratification. Increasing detection of malignancy with increasing age.

With regards to PSA at presentation, malignancy was detected in 15.3% of patients with PSA 0 to 10ng/ml, 16.2% with PSA 11 to 20ng/ml, 39% with PSA 21 to 50ng/ml and 77.9% with PSA >50ng/ml. Inflammation was seen in 13.8% of patients with PSA 0 to 10ng/ml, 27% with PSA 11 to 20ng/ml,24.7% with PSA 21 to 50ng/ml and 6.5% with PSA more than 50ng/ml. PIN was seen in 11% of patients with PSA 0 to 10ng/ml, 12.6% with PSA 11 to 20ng/ml, 3.9% with PSA 21 to 50ng/ml and 0% with PSA more than 50ng/ml (Figure 4)

As expected, patients diagnosed with malignancy had a higher median PSA of 26.6 +/- 230ng/ml as compared with benign disease (7.3 +/- 82ng/mL,) inflammation (11.7 +/- 23.7ng/mL) and PIN (8.2 +/- 7.1ng/mL). In our study, the PSA level did not show any correlation with prostate volume.

The higher the prostate volume, the lower the cancer detection. 60% of prostates weighing less than 20g were found to be malignant, 35.3% malignant for volume 20 to 29g, 24% for volume 30 to 39g, 21.4% for volume 40 to 49g and 14.5% for volume more than 50g. PIN was generally not seen in prostates less than 40g. Inflammation however, was seen in 14 to 17% of cases regardless of the prostate volume.

Of the 25.6% of patients with malignancy, 28% of them had a Gleason sum of 6, 24.8% with Gleason 7, 22.4% with Gleason 8, 24.2% with Gleason 9 and 0.6% with Gleason 10. Gleason sum 7 was seen in 30.8%, 19.3% and 28.6% of patients in their 6th, 7th and 8th decade of life respectively. Gleason sum 8 and above was seen in 60.6%, 43.9% and 43.1% of patients in their 6th,7th and 8th decade of life respectively. Patients above 80 years presented with significantly higher grade disease. 83.3% had gleason sum 8 and above.

Cancer Detection from Transrectal Ultrasound Guided Biopsy in a Single Center 99

The distribution of the number of cores taken was similar between patients diagnosed with malignancy and benign disease. 65.8% patients diagnosed with malignancy had 12 or more biopsies taken compared to 73.5% in patients diagnosed with benign disease. (p>0.05) Therefore in this analysis, the number of cores taken did not influence cancer detection rates. Wether the operator performing the biopsy is an independent factor predicting cancer detection rate was not studied in this analysis. Nathan et al 4suggested significant differences in operators performing transrectal biopsy in the detection of prostate cancer. It is evident that cancer detection was much better in the smaller prostates, especially when the volume was less than 20g. The diagnosis of malignancy decreases with increasing prostate volume. This is supported by many other studies including a study by Remzi et al 5 suggesting a repeat biopsy in prostates with total volume more than 20 mls with a negative first biopsy.This is probably due to easier detection in a smaller volume prostate undergoing biopsy. Reitbergen et al 6 found that most important factor responsible for the failure of diagnosis of prostate cancer at the primary screening was prostate volume. Another hypothesis is the higher probability of a smaller prostate with elevated PSA harbouring cancer as compared to a larger sized prostate gland. However, although Chinese patients generally presented with smaller prostates, their cancer detection rate was no higher than the other ethnic groups. It is also interesting to note that PIN was only identified in prostates more than 40g. Inflammation was observed at equal

As expected, the cancer detection rate increased with increasing PSA ranging from 15.3% detection for PSA 0 to 10ng/ml to 77.9% for PSA more than 50ng/ml. There was a significant rise in cancer detection with a PSA > 20ng/mL. Prostate inflammation detection was 14%-27% among patients with PSA< 50ng/mL but was significantly less (6.5%) in patients with PSA more than 50ng/ml. This finding does not suggest any correlation between prostate inflammation and malignancy in this cohort of patients . Terekawa et al 7 found an inverse relationship between histologic inflammation and prostate cancer in men

Detection of PIN was significantly higher in patients with PSA 0 to 20ng/ml compared to higher PSA levels. Within this range of PSA , 12% of biopsies had PIN . The incidence of isolated PIN in prostate biopsies varies in the literature. In urological practice, incidence

Cancer detection increased with increasing age averaging 30 to 50% in patients above 70 years old. Patients below 50 years who were biopsied did not have cancer and none had PIN, however 11.1% had prostate inflammation and the rest had benign disease. The lowest incidence of prostate inflammation was seen in patients older than 80 years. These patients

This study also showed the expected trend of a rising prostate volume with increasing age. From the age of 50 years onwards, there was a 10g increase in prostate volume in every decade and this trend stabilises after the 8th decade of life. Median PSA also notably increased with age. Men older than 80 years old had the highest median PSA, 26.5 ng/mL

From this preliminary study it appears that there could be some differences in the presentation of prostate disease and PSA distribution in Asian patients as compared to their western counterpart. The patients in this cohort were detected with malignancy at a higher PSA compared to their western counterparts. Further research would be required to explore these differences and to further study the PSA variation between the various ethnic groups.

and when diagnosed with malignancy had a significantly higher grade cancer.

rates in small and large prostates (14 to 17%).

with PSA 10-50 ng/mL undergoing prostate biopsy.

also had the highest cancer detection rate.

varies between 4.4-25% 8

Fig. 4. Line graph shows the probability of detecting various histological diagnosis based on range of PSA at presentation, it acts as a useful guide for the clinician who is doing the biopsy.

#### **5. Conclusion**

Cancer detection rate in our center was 25.6% and it appears to be the same among the 3 major ethnic groups. It was noticed that Malay patients presented with a higher median PSA and this is not explained by a large prostate volume as the Indian patients had similarly large prostates but presented with a lower PSA level. Prostate inflammation among Malay patients was similar with that of the Chinese patients. Therefore, it is possible that there may be other factors that could be contributing to the higher PSA among Malays . It was also interesting to note that despite presenting with a higher PSA level, Malay patients had similar cancer detection rate as the other two ethnic groups.

There appears to be variation in PSA among the various ethnic groups , however this is an area that will require further research as there may be other confounding factors that would contribute to the differences seen.

Fig. 4. Line graph shows the probability of detecting various histological diagnosis based on range of PSA at presentation, it acts as a useful guide for the clinician who is doing the

Cancer detection rate in our center was 25.6% and it appears to be the same among the 3 major ethnic groups. It was noticed that Malay patients presented with a higher median PSA and this is not explained by a large prostate volume as the Indian patients had similarly large prostates but presented with a lower PSA level. Prostate inflammation among Malay patients was similar with that of the Chinese patients. Therefore, it is possible that there may be other factors that could be contributing to the higher PSA among Malays . It was also interesting to note that despite presenting with a higher PSA level, Malay patients had

There appears to be variation in PSA among the various ethnic groups , however this is an area that will require further research as there may be other confounding factors that would

similar cancer detection rate as the other two ethnic groups.

contribute to the differences seen.

biopsy.

**5. Conclusion** 

The distribution of the number of cores taken was similar between patients diagnosed with malignancy and benign disease. 65.8% patients diagnosed with malignancy had 12 or more biopsies taken compared to 73.5% in patients diagnosed with benign disease. (p>0.05) Therefore in this analysis, the number of cores taken did not influence cancer detection rates.

Wether the operator performing the biopsy is an independent factor predicting cancer detection rate was not studied in this analysis. Nathan et al 4suggested significant differences in operators performing transrectal biopsy in the detection of prostate cancer.

It is evident that cancer detection was much better in the smaller prostates, especially when the volume was less than 20g. The diagnosis of malignancy decreases with increasing prostate volume. This is supported by many other studies including a study by Remzi et al 5 suggesting a repeat biopsy in prostates with total volume more than 20 mls with a negative first biopsy.This is probably due to easier detection in a smaller volume prostate undergoing biopsy. Reitbergen et al 6 found that most important factor responsible for the failure of diagnosis of prostate cancer at the primary screening was prostate volume. Another hypothesis is the higher probability of a smaller prostate with elevated PSA harbouring cancer as compared to a larger sized prostate gland. However, although Chinese patients generally presented with smaller prostates, their cancer detection rate was no higher than the other ethnic groups. It is also interesting to note that PIN was only identified in prostates more than 40g. Inflammation was observed at equal rates in small and large prostates (14 to 17%).

As expected, the cancer detection rate increased with increasing PSA ranging from 15.3% detection for PSA 0 to 10ng/ml to 77.9% for PSA more than 50ng/ml. There was a significant rise in cancer detection with a PSA > 20ng/mL. Prostate inflammation detection was 14%-27% among patients with PSA< 50ng/mL but was significantly less (6.5%) in patients with PSA more than 50ng/ml. This finding does not suggest any correlation between prostate inflammation and malignancy in this cohort of patients . Terekawa et al 7 found an inverse relationship between histologic inflammation and prostate cancer in men with PSA 10-50 ng/mL undergoing prostate biopsy.

Detection of PIN was significantly higher in patients with PSA 0 to 20ng/ml compared to higher PSA levels. Within this range of PSA , 12% of biopsies had PIN . The incidence of isolated PIN in prostate biopsies varies in the literature. In urological practice, incidence varies between 4.4-25% 8

Cancer detection increased with increasing age averaging 30 to 50% in patients above 70 years old. Patients below 50 years who were biopsied did not have cancer and none had PIN, however 11.1% had prostate inflammation and the rest had benign disease. The lowest incidence of prostate inflammation was seen in patients older than 80 years. These patients also had the highest cancer detection rate.

This study also showed the expected trend of a rising prostate volume with increasing age. From the age of 50 years onwards, there was a 10g increase in prostate volume in every decade and this trend stabilises after the 8th decade of life. Median PSA also notably increased with age. Men older than 80 years old had the highest median PSA, 26.5 ng/mL and when diagnosed with malignancy had a significantly higher grade cancer.

From this preliminary study it appears that there could be some differences in the presentation of prostate disease and PSA distribution in Asian patients as compared to their western counterpart. The patients in this cohort were detected with malignancy at a higher PSA compared to their western counterparts. Further research would be required to explore these differences and to further study the PSA variation between the various ethnic groups.

**7** 

Κ. Stamatiou

*Greece* 

**Elderly and Early Prostate Cancer** 

Prostate cancer (PC) is the second most frequent malignant disease and the second-leading cause of cancer deaths among men in the United States (1). Both evidence and epidemiologic studies have shown that PC is rare in men younger than 50 years of age, but thereafter the risk of incident prostate cancer increases significantly with increasing age (2). After the introduction and widespread use of the prostate-specific antigen (PSA) blood test, PC incidence has increased and it is expected that this disease is likely to become a more prominent and pressing problem in many countries as the percentage of elderly men increases (3). Actually declines in mortality at younger ages, medical advances, and better health care have resulted in longer life expectancy in both the developing and the developed world the last three decades. Statistics compiled by the United Nations showed that in 1999, 10% of the world population was 60 years and older (4). By 2050, this percentage will rise to 22%. In Hong Kong, where the proportion of elderly is even higher, it is estimated to rise to 40%. Regarding male gender, the population over 65 years is expected to increase 4-fold worldwide by 2050 (5). Achievements of the 20th century have changed the world's demographic proportions without altering the epidemiology of PC. Indeed, PC still remains a disease of elderly men and thus, increased PC incidence could be partly attributed to the

While the majority of elderly PC patients in the past were diagnosed with local advanced or metastatic disease, a rising number of elderly men are now diagnosed with early stage PC. It is not known whether and if this is due to the effective utilization of health care resources or to the widespread use of PSA testing worldwide. Several studies however showed that after the introduction of PSA, additionally to the increase in the PC detection rate, an eventual shift towards earlier pathological stage was occurred also (3). The increased life expectancy enjoyed by the world population also means that the life span beyond age 60 is much longer than demographers have previously envisaged. Currently, a large proportion of the population remains active beyond the age of 70 and lives beyond the age of 80. Since many of them are healthy, the number of elderly men who will be diagnosed with PC and may

The aim of present study is to discuss the issue of screening for PC in elderly individuals as well as to review the current data on the treatment of early stage PC in elder males. A secondary aim is to examine whether or not advanced age impacts on PC risk. The impact of life expectancy on the choice of treatment in both patients and health care providers has

**1. Introduction** 

steadily growing ageing population (6).

been investigated also.

require treatment will further increase in the coming years (7).

*Urology department Tzaneion General Hospital of Pireas* 

#### **6. Acknowledgement**

Dr Teo Swi Han, Dr Nurul Hayati bt Abu Hassan and Dr Khalid bin Othman for their effort in tracing all the patient records.

Keywords: TRUS, Prostate Cancer, Inflammation

#### **7. References**


### **Elderly and Early Prostate Cancer**

Κ. Stamatiou

*Urology department Tzaneion General Hospital of Pireas Greece* 

#### **1. Introduction**

100 Prostate Cancer – Original Scientific Reports and Case Studies

Dr Teo Swi Han, Dr Nurul Hayati bt Abu Hassan and Dr Khalid bin Othman for their effort

[2] Sothilingam S et al 2010 , *Prostate Cancer Screening, Malaysian Perspective*, Urologic

[3] Chia et al , 2007, *PSA levels among Chinese, Malays and Indians in Singapore from a* 

[4] Nathan et al , 2009, *Operator is an Independent Predictor of Detecting Prostate Cancer at* 

[5] Remzi et al, 2003, *Can Total and Transition Zone Volume of Prostate determine whether to* 

[6] Reietbergen et al , 1998 , *Repeat screening for prostate cancer after 1 year followup in 984 biopsied men*: Clinical and pathological factors in detecting cancer., J Urol: 160-2121-5. [7] Terekawa et al , 2008, *Inverse Association between Histologic Inflammation in Needle Biopsy* 

[8] Jonian et al , 2005,*Prostatic Intraepithelial Neoplasia (PIN): Importance and Clinical* 

*specimens and Prostate Cancer in men with serum PSA 10-50 ng/mL,* , Urology: 1194-1197.

*community based study.* Asian Pac J Cancer Prev Jul-Sept; 8(3): 375-8

*Transrectal Ultrasound Guided Prostate Biopsy*., J Urol: 182, 2659-2663.

Oncology- Seminars and Original Investigations; 28 670-72

**6. Acknowledgement** 

**7. References** 

in tracing all the patient records.

Keywords: TRUS, Prostate Cancer, Inflammation

[1] Lim TC, Norraha AR, 2006, *Natioanal Cancer Registry*

*perform a repeat biopsy*, Urology; 61(1): 161-6

*Management* . European Urology 48 , 379-385.

Prostate cancer (PC) is the second most frequent malignant disease and the second-leading cause of cancer deaths among men in the United States (1). Both evidence and epidemiologic studies have shown that PC is rare in men younger than 50 years of age, but thereafter the risk of incident prostate cancer increases significantly with increasing age (2). After the introduction and widespread use of the prostate-specific antigen (PSA) blood test, PC incidence has increased and it is expected that this disease is likely to become a more prominent and pressing problem in many countries as the percentage of elderly men increases (3). Actually declines in mortality at younger ages, medical advances, and better health care have resulted in longer life expectancy in both the developing and the developed world the last three decades. Statistics compiled by the United Nations showed that in 1999, 10% of the world population was 60 years and older (4). By 2050, this percentage will rise to 22%. In Hong Kong, where the proportion of elderly is even higher, it is estimated to rise to 40%. Regarding male gender, the population over 65 years is expected to increase 4-fold worldwide by 2050 (5). Achievements of the 20th century have changed the world's demographic proportions without altering the epidemiology of PC. Indeed, PC still remains a disease of elderly men and thus, increased PC incidence could be partly attributed to the steadily growing ageing population (6).

While the majority of elderly PC patients in the past were diagnosed with local advanced or metastatic disease, a rising number of elderly men are now diagnosed with early stage PC. It is not known whether and if this is due to the effective utilization of health care resources or to the widespread use of PSA testing worldwide. Several studies however showed that after the introduction of PSA, additionally to the increase in the PC detection rate, an eventual shift towards earlier pathological stage was occurred also (3). The increased life expectancy enjoyed by the world population also means that the life span beyond age 60 is much longer than demographers have previously envisaged. Currently, a large proportion of the population remains active beyond the age of 70 and lives beyond the age of 80. Since many of them are healthy, the number of elderly men who will be diagnosed with PC and may require treatment will further increase in the coming years (7).

The aim of present study is to discuss the issue of screening for PC in elderly individuals as well as to review the current data on the treatment of early stage PC in elder males. A secondary aim is to examine whether or not advanced age impacts on PC risk. The impact of life expectancy on the choice of treatment in both patients and health care providers has been investigated also.

Elderly and Early Prostate Cancer 103

influences prostate cancer growth, expansion of CAG repeats in the AR affect the risk of

Age-dependent clonal transformation events that affect the risk of developing prostate cancer may also occur to the Insulin-like Growth Factor-II (IGF-II) gene. The IGF-II gene is an auto- paracrine growth stimulator that is an important positive modulator of cancer development. IGF-II losses of imprinting, as well as increased lGF-II expression resulting from age-dependent changes in DNA methylation, have been recently associated with

A major consideration for cancer screening is to weigh up the possibility someone will have needless treatment against saving lives. PC can develop into a fatal, painful disease, but it can also develop so slowly that it will never cause problems during the man's lifetime. Actually, although none of the existing screening tools can accurately distinguish between lethal and indolent PC, the use of PSA has been shown to increase the PC detection rate with a shift to detection at earlier and therefore curable stages (12). This fact generated also concerns about over-diagnosis and over-treating and arguments both for and against the efficacy of screening. Under the light of this evidence it became clear why the issues of overdiagnosis and over-treating are of outmost importance when deciding to screen elderly

Data from US Cancer of the Prostate Strategic Urological Research Endeavor shows that most of the patients diagnosed with prostate cancer the last two decades in the US had low or intermediate disease at diagnosis (14). Moreover, between 20 and 30% of PCs found in radical prostatectomy specimens of men with PSA-detected disease are non palpable, potentially indolent cancers (Gleason <6, tumor volume <0,5cm3) (15,16). Since doubling time of high and intermediate differentiated prostatic carcinomas reaches 7 and 5 years respectively, a small tumor (<0,5cm3) poses little threat for the life of older individuals (from the perspective that needs enough time to became life threatening). In confirmation to the above, Albertsen and colleagues demonstrated that men with prostate biopsy specimens showing Gleason score 2 to 4 disease faced a minimal risk of death from prostate cancer within 15 years from diagnosis (17). Given that life expectancy of American males at the age of 65 is 16 years (18) and the mean time to cancer-specific death of apparently clinically localized prostate cancer is 17 years (19), it became obvious why PC screening and treatment of PSA detected PCs in elderly patients is a controversial issue. Most doctors however argue against PSA testing for men who are in their 70s or older, because even if prostate cancer were detected, most men would be dead of something else before the cancer progressed (20). This is true only in part. As previously mentioned, today a large proportion of the population lives beyond age 70 and many are healthy. These men have several reasons –the belief in the benefit of early diagnosis, the need to have trust, and a desire for reliable screening resembling women- to undergo testing for prostate cancer. Yet, patient's anxiety increase the likelihood of getting the screening test, by acting powerfully on the screening decisions of physicians, whose clinical judgment would otherwise make them least, inclined

At the moment, PC screening is being performed unofficially in elderly patients visiting outpatient departments of general hospitals and consulting rooms. The exact magnitude of this opportunistic screening is not known however it is believed that reaches high numbers

developing prostate cancer in a race and age depending matter (10).

**3.2 The issue of screening for prostate cancer in elderly individuals** 

increased risk for PC development (11).

individuals (13).

to order the test (21).

#### **2. Methods**

We identified studies published from 1990 onwards by searching the MEDLINE database of the National Library of Medicine. Initial search terms were localized prostate cancer, early stage prostate cancer, combined with elderly patients, life expectancy, palliative, curative, quality of life, watchful waiting, radical prostatectomy, brachytherapy and external beam radiotherapy. References in the selected publications were checked for relevant publications not included in the Medline/Pubmed search.

#### **3. Results**

#### **3.1 How ageing can increase the risk of prostate cancer?**

A definitive cause of PC has not been identified and the specific mechanisms that lead to the development of the disease are still unknown. Although several risk factors have been proposed, the only risk factors that can be considered established are age, race and family history. Evidence suggests that an association between the above risk factors through a common pathogenic mechanism exists. On one hand, development and function of the prostate gland is endocrine controlled and androgen/estrogen synergism is necessary for the integrity of the normal human prostate. On the other hand androgen action is critical to the development, progression and cure of PC. Under those circumstances, it could be expected that ageing facilitate PC development through androgenic action. In fact, androgens undergo a significant age-dependent alteration: with ageing the production of testosterone by the testes is decreasing leading thus in a significant reduction of the endogenous testosterone levels. DHT activity decreases in the epithelium while in the stroma it remains constant over the whole age range. The age-dependent decrease of the DHT accumulation in epithelium and the concomitant increase of the estrogen accumulation in stroma lead to a tremendous increase of the estrogen/androgen ratio in the human prostate. Although, the specific pathway remains partially investigated, it is widely accepted that these alterations promote the initiation of benign prostatic hypertrophy, the most common disease of the ageing prostate. Similarly to benign prostatic hypertrophy, PC incidence increases with age: it seldom develops before the age of 40 and is chiefly a disease found in men over the age 65 years. Epidemiological evidence from autopsy studies show that while a very high proportion of elderly men has histological evidence of the disease, a much smaller proportion actually develop clinically apparent PC however, most of the impalpable cancers likely to progress and become clinically significant (advanced Gleason score, greater volume) are found in older individuals (8). However, age-related increase in the prevalence of prostate cancer found in autopsy is not similar worldwide. Variations in the reported incidence of PC between different racial groups suggest that some populations are either more susceptible to PC-promoting events or are exposed to different promoting agents (9). Take the above in consideration it could be speculated that ageing may promote clonal transformation events of pathogenetic importance for the initiation of PC. These clonal transformation events may be boosted by genetic predisposing factors. Although the exact pathways remain unknown, evidence suggests that they involving the androgen receptor (AR). The AR is a structurally conserved member of the nuclear receptor superfamily and signalling via the AR is critical for carcinogenesis and progression of the disease. The AR´s amino-terminal domain is required for transcriptional activation and contains a region of polyglutamine encoded by CAG trinucleotide repeats. As androgen

We identified studies published from 1990 onwards by searching the MEDLINE database of the National Library of Medicine. Initial search terms were localized prostate cancer, early stage prostate cancer, combined with elderly patients, life expectancy, palliative, curative, quality of life, watchful waiting, radical prostatectomy, brachytherapy and external beam radiotherapy. References in the selected publications were checked for relevant publications

A definitive cause of PC has not been identified and the specific mechanisms that lead to the development of the disease are still unknown. Although several risk factors have been proposed, the only risk factors that can be considered established are age, race and family history. Evidence suggests that an association between the above risk factors through a common pathogenic mechanism exists. On one hand, development and function of the prostate gland is endocrine controlled and androgen/estrogen synergism is necessary for the integrity of the normal human prostate. On the other hand androgen action is critical to the development, progression and cure of PC. Under those circumstances, it could be expected that ageing facilitate PC development through androgenic action. In fact, androgens undergo a significant age-dependent alteration: with ageing the production of testosterone by the testes is decreasing leading thus in a significant reduction of the endogenous testosterone levels. DHT activity decreases in the epithelium while in the stroma it remains constant over the whole age range. The age-dependent decrease of the DHT accumulation in epithelium and the concomitant increase of the estrogen accumulation in stroma lead to a tremendous increase of the estrogen/androgen ratio in the human prostate. Although, the specific pathway remains partially investigated, it is widely accepted that these alterations promote the initiation of benign prostatic hypertrophy, the most common disease of the ageing prostate. Similarly to benign prostatic hypertrophy, PC incidence increases with age: it seldom develops before the age of 40 and is chiefly a disease found in men over the age 65 years. Epidemiological evidence from autopsy studies show that while a very high proportion of elderly men has histological evidence of the disease, a much smaller proportion actually develop clinically apparent PC however, most of the impalpable cancers likely to progress and become clinically significant (advanced Gleason score, greater volume) are found in older individuals (8). However, age-related increase in the prevalence of prostate cancer found in autopsy is not similar worldwide. Variations in the reported incidence of PC between different racial groups suggest that some populations are either more susceptible to PC-promoting events or are exposed to different promoting agents (9). Take the above in consideration it could be speculated that ageing may promote clonal transformation events of pathogenetic importance for the initiation of PC. These clonal transformation events may be boosted by genetic predisposing factors. Although the exact pathways remain unknown, evidence suggests that they involving the androgen receptor (AR). The AR is a structurally conserved member of the nuclear receptor superfamily and signalling via the AR is critical for carcinogenesis and progression of the disease. The AR´s amino-terminal domain is required for transcriptional activation and contains a region of polyglutamine encoded by CAG trinucleotide repeats. As androgen

**2. Methods** 

**3. Results** 

not included in the Medline/Pubmed search.

**3.1 How ageing can increase the risk of prostate cancer?** 

influences prostate cancer growth, expansion of CAG repeats in the AR affect the risk of developing prostate cancer in a race and age depending matter (10).

Age-dependent clonal transformation events that affect the risk of developing prostate cancer may also occur to the Insulin-like Growth Factor-II (IGF-II) gene. The IGF-II gene is an auto- paracrine growth stimulator that is an important positive modulator of cancer development. IGF-II losses of imprinting, as well as increased lGF-II expression resulting from age-dependent changes in DNA methylation, have been recently associated with increased risk for PC development (11).

#### **3.2 The issue of screening for prostate cancer in elderly individuals**

A major consideration for cancer screening is to weigh up the possibility someone will have needless treatment against saving lives. PC can develop into a fatal, painful disease, but it can also develop so slowly that it will never cause problems during the man's lifetime. Actually, although none of the existing screening tools can accurately distinguish between lethal and indolent PC, the use of PSA has been shown to increase the PC detection rate with a shift to detection at earlier and therefore curable stages (12). This fact generated also concerns about over-diagnosis and over-treating and arguments both for and against the efficacy of screening. Under the light of this evidence it became clear why the issues of overdiagnosis and over-treating are of outmost importance when deciding to screen elderly individuals (13).

Data from US Cancer of the Prostate Strategic Urological Research Endeavor shows that most of the patients diagnosed with prostate cancer the last two decades in the US had low or intermediate disease at diagnosis (14). Moreover, between 20 and 30% of PCs found in radical prostatectomy specimens of men with PSA-detected disease are non palpable, potentially indolent cancers (Gleason <6, tumor volume <0,5cm3) (15,16). Since doubling time of high and intermediate differentiated prostatic carcinomas reaches 7 and 5 years respectively, a small tumor (<0,5cm3) poses little threat for the life of older individuals (from the perspective that needs enough time to became life threatening). In confirmation to the above, Albertsen and colleagues demonstrated that men with prostate biopsy specimens showing Gleason score 2 to 4 disease faced a minimal risk of death from prostate cancer within 15 years from diagnosis (17). Given that life expectancy of American males at the age of 65 is 16 years (18) and the mean time to cancer-specific death of apparently clinically localized prostate cancer is 17 years (19), it became obvious why PC screening and treatment of PSA detected PCs in elderly patients is a controversial issue. Most doctors however argue against PSA testing for men who are in their 70s or older, because even if prostate cancer were detected, most men would be dead of something else before the cancer progressed (20).

This is true only in part. As previously mentioned, today a large proportion of the population lives beyond age 70 and many are healthy. These men have several reasons –the belief in the benefit of early diagnosis, the need to have trust, and a desire for reliable screening resembling women- to undergo testing for prostate cancer. Yet, patient's anxiety increase the likelihood of getting the screening test, by acting powerfully on the screening decisions of physicians, whose clinical judgment would otherwise make them least, inclined to order the test (21).

At the moment, PC screening is being performed unofficially in elderly patients visiting outpatient departments of general hospitals and consulting rooms. The exact magnitude of this opportunistic screening is not known however it is believed that reaches high numbers

Elderly and Early Prostate Cancer 105

and risks of diagnostic procedures and treatment be taken into account when considering whether to undertake PC screening. On the other hand, treatment recommendations are now recognizing that older men with PC should be managed according to their individual health status, which is mainly driven by the severity of associated comorbid conditions, and not according to chronological age. According to the International Society of Geriatric Oncology Prostate Cancer Task Force, it is possible, based on a rapid and simple evaluation, to classify patients into four different groups: 1) "Healthy" patients (controlled comorbidity, fully independent in daily living activities, and no malnutrition) should receive the same treatment as younger patients; 2) "Vulnerable" patients (reversible impairment) should receive standard treatment after medical intervention; 3) "Frail" patients (irreversible impairment) should receive adapted treatment; 4) Patients who are "too sick" with "terminal illness" should receive only symptomatic palliative treatment.(30) The same rapid and simple evaluation may help physicians who perform

The main treatment options include radical prostatectomy (RP), radiotherapy (external beam radiotherapy and brachytherapy), watchful waiting (WW) and androgen deprivation therapy (ADT). Other include, cryotherapy (freezing the prostate), high-intensity focused ultrasound (HIFU), radiofrequency interstitial tumour ablation (RITA) and non-hormonal therapy (cytotoxic agents). Radical prostatectomy, brachytherapy and external beam radiotherapy are considered curative, while watchful waiting and hormone-therapy palliative. All treatments have risks of complications, although frequency and severity may vary. The primary goal of treatment is to target the men most likely to need intervention in order to prevent prostate cancer death and disability while minimizing intervention-related complications. However, whereas the standard oncologic evaluation works reasonably well in most other populations, in elderly PC patients, tends to overestimate possible harms associated with radical treatment and underestimate patients ability to withstand treatments side effects. In accordance to the above, various studies have demonstrated that potentially curative therapy (radical prostatectomy or radiotherapy) is applied less often in older PC patients (32,33,34,35,36). Traditionally, PC is considering a slow progressive disease that needs enough time to become life threatening for an elder individual and this possibly explains the above observation. However, a multivariate analysis of the SEER database revealed significantly decreased odds of receiving cancer directed surgery in the elderly patient with lung, liver, breast, pancreas, esophageal, gastric cancers, sarcoma and rectal cancer while other studies have demonstrated under use of cancer directed radiation and chemotherapy (37,38,39). These findings are posing justifiable concern about under-treatment of the elderly cancer patient and raise the provocative question if this is due to judicious, evidence based selection or discrimination based only on age (40). The reasons for the observed under use of cancer directed treatment in the elderly remain elusive. However, discrimination -if presentreflects the stereotypes that older people are physically frail, unfit for curative treatment, indisposed to accept treatment related complications, impatient and uninterested in prolonging survival. With regard to PC treatment decision making, increasing age is definitely a risk factor for receiving inadequate treatment (41). Harlan et al demonstrated that advantaged age is- still- considering as important as PSA, clinical stage and Gleason score while other demonstrated that age is the predominant factor influencing treatment

PSA screening to decide who to screen.

**3.3 Treatment options and treatment decision making** 

worldwide. Hoffman and associates and Walter and colleagues found a 56% and 50% PSA screening rate in their cohort of elderly men in 2003 and 2010, respectively (22,23). Bowen and co-workers found that PC screening rates among men at the age of 80 and older are even higher than that of men in the age range of 50 to 64 years (64% versus 56%)(24). Similarly, in a study by D'Ambrosio and colleagues, the highest yearly exposure to PSA screening and the highest frequency of repeat testing were observed in the age range of 70 to the 79 years (25). In contrast, Zeliadt and associates demonstrated that PSA testing among men older than 75 years has declined slightly following the recommendations by the US Preventive Services Task Force in 2008 and is still continuing to decline (15). Aus and colleagues found that restrictions in the use of PSA test in individuals over 75 years resulted in PC incidence falls after peaking at the age of 75 (26). Interestingly, evidence suggests that PSA testing may be useful in diagnosis of aggressive early PC in a subset of elderly patients. A current study by Brassell and colleagues showed that as men age, parameters consistent with more aggressive disease become more prevalent (27) a fact that was confirmed by the findings of an autopsy study demonstrating that a proportion of elderly men with histologically apparent disease may develop lethal PC (21). Therefore it is not surprising that older individuals with clinically apparent PC usually die from PC. These data may have implications for future screening and treatment recommendations. Currently, age plays an important role in both screening decision and treatment choice and thus elderly patients are less likely to undergo PSA test and receive local therapy.

Data from US Cancer of the Prostate Strategic Urological Research Endeavor shows a significant reduction of risk of death from metastatic prostate cancer and a decrease in prostate carcinoma-specific mortality the last two decades in the US (14). As yet it is not possible to say what proportion of the fall in mortality is the result of improvements in treatment, changes in cancer registration coding, the attribution of death to PC, and the effects of PSA testing. Accumulative evidence however suggests that early screening of PC in asymptomatic men reduce their risk of death from metastatic disease. Interestingly, the recently published results of the European Randomised Study for Screening of Prostate Cancer (ERSPC) reported a relative PC mortality reduction of at least 20% by PSA-based population screening (28) while Goel and Kopec reported an even higher reduction of risk of death from metastatic PC among men who were not screened regularly as part of a screening program (29).

Given that PSA screening mainly diagnoses early PC, it may be justifiable for otherwise healthy elderly men to undergo PSA test. This is of outmost importance since older patients are more likely to have high-risk prostate cancer at diagnosis and lower overall survival. In fact, under-use of potentially curative local therapy among older men with high-risk disease may explain, at least in part, the observed differences in cancer-specific survival across age strata.(30). Taking in consideration these findings along with observations of Brassell and coworkers (27) it became obvious that evidence supports making decisions regarding screening on the basis of disease risk and life expectancy rather than chronologic age.

Currently, no standard recommendation for PC screening exists. Recently, the American Urological Association recommends PC screening to men aged 40 years or older. In contrast, screening is presently discouraged by the EC Advisory Committee on Cancer Prevention for its negative effects are evident and its benefits still uncertain (31). According to the U.S. Preventive Services Task Force, evidence is insufficient to recommend in favour of, or against routine PC screening (23). The abovementioned professional organizations and health agencies as well as most of medical experts agree that it is important that the benefits

worldwide. Hoffman and associates and Walter and colleagues found a 56% and 50% PSA screening rate in their cohort of elderly men in 2003 and 2010, respectively (22,23). Bowen and co-workers found that PC screening rates among men at the age of 80 and older are even higher than that of men in the age range of 50 to 64 years (64% versus 56%)(24). Similarly, in a study by D'Ambrosio and colleagues, the highest yearly exposure to PSA screening and the highest frequency of repeat testing were observed in the age range of 70 to the 79 years (25). In contrast, Zeliadt and associates demonstrated that PSA testing among men older than 75 years has declined slightly following the recommendations by the US Preventive Services Task Force in 2008 and is still continuing to decline (15). Aus and colleagues found that restrictions in the use of PSA test in individuals over 75 years resulted in PC incidence falls after peaking at the age of 75 (26). Interestingly, evidence suggests that PSA testing may be useful in diagnosis of aggressive early PC in a subset of elderly patients. A current study by Brassell and colleagues showed that as men age, parameters consistent with more aggressive disease become more prevalent (27) a fact that was confirmed by the findings of an autopsy study demonstrating that a proportion of elderly men with histologically apparent disease may develop lethal PC (21). Therefore it is not surprising that older individuals with clinically apparent PC usually die from PC. These data may have implications for future screening and treatment recommendations. Currently, age plays an important role in both screening decision and treatment choice and thus elderly patients are

Data from US Cancer of the Prostate Strategic Urological Research Endeavor shows a significant reduction of risk of death from metastatic prostate cancer and a decrease in prostate carcinoma-specific mortality the last two decades in the US (14). As yet it is not possible to say what proportion of the fall in mortality is the result of improvements in treatment, changes in cancer registration coding, the attribution of death to PC, and the effects of PSA testing. Accumulative evidence however suggests that early screening of PC in asymptomatic men reduce their risk of death from metastatic disease. Interestingly, the recently published results of the European Randomised Study for Screening of Prostate Cancer (ERSPC) reported a relative PC mortality reduction of at least 20% by PSA-based population screening (28) while Goel and Kopec reported an even higher reduction of risk of death from metastatic PC among men who were not screened regularly as part of a

Given that PSA screening mainly diagnoses early PC, it may be justifiable for otherwise healthy elderly men to undergo PSA test. This is of outmost importance since older patients are more likely to have high-risk prostate cancer at diagnosis and lower overall survival. In fact, under-use of potentially curative local therapy among older men with high-risk disease may explain, at least in part, the observed differences in cancer-specific survival across age strata.(30). Taking in consideration these findings along with observations of Brassell and coworkers (27) it became obvious that evidence supports making decisions regarding screening on the basis of disease risk and life expectancy rather than chronologic age. Currently, no standard recommendation for PC screening exists. Recently, the American Urological Association recommends PC screening to men aged 40 years or older. In contrast, screening is presently discouraged by the EC Advisory Committee on Cancer Prevention for its negative effects are evident and its benefits still uncertain (31). According to the U.S. Preventive Services Task Force, evidence is insufficient to recommend in favour of, or against routine PC screening (23). The abovementioned professional organizations and health agencies as well as most of medical experts agree that it is important that the benefits

less likely to undergo PSA test and receive local therapy.

screening program (29).

and risks of diagnostic procedures and treatment be taken into account when considering whether to undertake PC screening. On the other hand, treatment recommendations are now recognizing that older men with PC should be managed according to their individual health status, which is mainly driven by the severity of associated comorbid conditions, and not according to chronological age. According to the International Society of Geriatric Oncology Prostate Cancer Task Force, it is possible, based on a rapid and simple evaluation, to classify patients into four different groups: 1) "Healthy" patients (controlled comorbidity, fully independent in daily living activities, and no malnutrition) should receive the same treatment as younger patients; 2) "Vulnerable" patients (reversible impairment) should receive standard treatment after medical intervention; 3) "Frail" patients (irreversible impairment) should receive adapted treatment; 4) Patients who are "too sick" with "terminal illness" should receive only symptomatic palliative treatment.(30) The same rapid and simple evaluation may help physicians who perform PSA screening to decide who to screen.

#### **3.3 Treatment options and treatment decision making**

The main treatment options include radical prostatectomy (RP), radiotherapy (external beam radiotherapy and brachytherapy), watchful waiting (WW) and androgen deprivation therapy (ADT). Other include, cryotherapy (freezing the prostate), high-intensity focused ultrasound (HIFU), radiofrequency interstitial tumour ablation (RITA) and non-hormonal therapy (cytotoxic agents). Radical prostatectomy, brachytherapy and external beam radiotherapy are considered curative, while watchful waiting and hormone-therapy palliative. All treatments have risks of complications, although frequency and severity may vary. The primary goal of treatment is to target the men most likely to need intervention in order to prevent prostate cancer death and disability while minimizing intervention-related complications. However, whereas the standard oncologic evaluation works reasonably well in most other populations, in elderly PC patients, tends to overestimate possible harms associated with radical treatment and underestimate patients ability to withstand treatments side effects. In accordance to the above, various studies have demonstrated that potentially curative therapy (radical prostatectomy or radiotherapy) is applied less often in older PC patients (32,33,34,35,36). Traditionally, PC is considering a slow progressive disease that needs enough time to become life threatening for an elder individual and this possibly explains the above observation. However, a multivariate analysis of the SEER database revealed significantly decreased odds of receiving cancer directed surgery in the elderly patient with lung, liver, breast, pancreas, esophageal, gastric cancers, sarcoma and rectal cancer while other studies have demonstrated under use of cancer directed radiation and chemotherapy (37,38,39). These findings are posing justifiable concern about under-treatment of the elderly cancer patient and raise the provocative question if this is due to judicious, evidence based selection or discrimination based only on age (40). The reasons for the observed under use of cancer directed treatment in the elderly remain elusive. However, discrimination -if presentreflects the stereotypes that older people are physically frail, unfit for curative treatment, indisposed to accept treatment related complications, impatient and uninterested in prolonging survival. With regard to PC treatment decision making, increasing age is definitely a risk factor for receiving inadequate treatment (41). Harlan et al demonstrated that advantaged age is- still- considering as important as PSA, clinical stage and Gleason score while other demonstrated that age is the predominant factor influencing treatment

Elderly and Early Prostate Cancer 107

AGE DEATHS % RATE 15-19 0 0 20-24 1 0,0 25-29 2 0,0 30-34 1 0,0 35-39 4 0,0 40-44 20 0,1 45-49 71 0,3 50-54 233 0,9 55-59 722 2,8 60-64 1738 6,9 65-69 3123 12,5 70-74 4636 18,5 75-79 5337 21,4 80-84 4536 18,2 > 85 4625 18,1

that of elderly PC patients treated with watchful waiting (50). Results from other studies showed that surgical therapy can achieve excellent oncologic results in selected elderly patients but they didn't found significant differences in overall survival (51,52). According to the results of the Scandinavian Prostate Cancer Group (SPCG) study, radical prostatectomy is associated with less deaths from prostate cancer (10 vs.15%), less deaths from any other cause (24 vs. 30%) and less metastases (14 vs. 23%) in a median follow-up of 8.2 years. However, benefit in cancer specific survival is limited to patients younger than 65 years. On the other hand there are several facts supporting WW (deferring intervention until the advent of symptoms) as an ideal treatment of early PC in elderly patients: Epidemiological evidence from autopsy studies show that while a very high proportion of elderly men has histological evidence of the disease, a much smaller proportion actually develop clinically apparent PC (53). Several authors demonstrated that elderly patients with localized PC have a favorable outlook following WW (54,55) and other showed that WW results in similar overall survival when compared with RP (56). In a pooled analysis of 828 case records from six nonrandomized studies, of men treated conservatively for clinically localized prostate cancer, Chodak et al, found an impressive 87% five years disease-specific survival rate (57), however other found that disease specific survival is better in patients who had undergone surgery and some authors argue that WW simply postpone the final treatment (58,59,60,61). Notably, there are no randomized clinical trials comparing surgery with radiation therapy in elderly PC patients, however an observational study, by Albertsen and colleagues showed that surgery is superior to radiation in localized prostate cancer in terms of

The truth is that the preferred management of clinically localized prostate cancer is not known, due in large part to the paucity of randomized controlled trials comparing the effectiveness and harms across primary treatment options. It seems that age itself is the main determinant of treatment selection: according to the Swedish Cancer Register in men with localized tumors expectant treatment was much more commonly used in those aged > or =75 years than in those aged <75 years (63). It is also clear that WW is an adequate

Table 1. Age and PC specific mortality (US Public service 1989).

prolonging overall and disease-specific survival (62).

decision making: Alibhai and colleagues generated an age-stratified random sample of 347 men from a cohort of patients with newly diagnosed prostate carcinoma in the Ontario Cancer Registry. Patients who were younger than 60 years were more likely to receive radical prostatectomy than radiation therapy or no therapy. Men between 60 and 69 years of age were more likely to receive radiation therapy than radical prostatectomy. Men between 70 and 79 years were most likely to receive no therapy, and nearly all men over 80 years received no therapy (42). Basically, although age plays a key role in treatment decision making, age itself is not predictive of outcome in an elderly cancer patient. In contrast, treatment outcome is strictly associated with clinical parameters such as the tumor stage, tumor grade or Gleason score and PSA level and therefore, treatment selection should be balanced between clinical stage and remaining life expectancy. It should be noticed however that the 10-year rule currently used to estimate life expectancy in elderly PC patients has demonstrated limited predictive validity and its use in clinical decision-making doesn't decrease the likelihood of receiving inappropriate treatment in elderly individuals (31).

Regarding localized PC, available treatment options include established therapies such as WW, RP, brachytherapy and external beam radiotherapy and non established therapies such as minimal invasive techniques and early hormone-therapy (43). RP is considering the gold standard for the treatment of localized PC and in fact it is the most common treatment with approximately 60,000 operations performed annually in the US (44). However, only a small number of elderly patients with early stage PC are treated with RP (45). The reason why advanced age is an unfavorable predictor of the probability of surgical treatment is not known (46). Actually, the fact that elderly individuals have lower life expectancy as well as the belief that elderly patients with localized disease are considering more prone to die with PC than of it, partly explain why PR is the less popular treatment of early PC in elderly patients. Moreover, elderly patients are often being considering fragile enough to receive surgical treatment. Whether and if age increase surgical risk is a controversial issue and for this reason several investigators claim that it is co-morbidity that actually increases the surgical risk and not ageing itself (47). Although, co-morbid illness has demonstrated increasing importance as a prognostic factor, its role is poorly defined. It is generally accepted that co-morbidity limits the generalization of results to older and sicker patients however; the widespread integration of co-morbidity into clinical practice has yet to be realized.

Reported differences in PC specific survival across age strata may be associated with under use of potentially curative local therapy among older men. In fact, PC mortality increases with ageing, peaks at the age of 70-75 and no significant decrease occurs thereafter (table 1). According to the SEER database, younger men (under age 65) with localized prostate cancer had 25-year prostate cancer mortality rates of approximately 19% for Gleason 6 disease, 37% for Gleason 7 disease, and 50% for Gleason 8-10 disease (48). Given that the survival advantage of surgery is most pronounced in men with higher stage disease, it became obvious that elderly PC patients with aggressive disease and life expectancy >10 years are likely to die from progressive prostate cancer (49). Worth mentioning, Bechis et al. studied men in the Cancer of the Prostate Strategic Urologic Research Endeavor (CaPSURE) database with complete risk, treatment, and follow-up information. They found that older patients are more likely to have high-risk prostate cancer at diagnosis and less likely to receive local therapy (30). In confirmation to the above, Dahm et al.showed that risk of death from PC for elderly PC patients treated with PR is significantly lower when compared with

decision making: Alibhai and colleagues generated an age-stratified random sample of 347 men from a cohort of patients with newly diagnosed prostate carcinoma in the Ontario Cancer Registry. Patients who were younger than 60 years were more likely to receive radical prostatectomy than radiation therapy or no therapy. Men between 60 and 69 years of age were more likely to receive radiation therapy than radical prostatectomy. Men between 70 and 79 years were most likely to receive no therapy, and nearly all men over 80 years received no therapy (42). Basically, although age plays a key role in treatment decision making, age itself is not predictive of outcome in an elderly cancer patient. In contrast, treatment outcome is strictly associated with clinical parameters such as the tumor stage, tumor grade or Gleason score and PSA level and therefore, treatment selection should be balanced between clinical stage and remaining life expectancy. It should be noticed however that the 10-year rule currently used to estimate life expectancy in elderly PC patients has demonstrated limited predictive validity and its use in clinical decision-making doesn't decrease the likelihood of receiving inappropriate treatment in elderly individuals (31). Regarding localized PC, available treatment options include established therapies such as WW, RP, brachytherapy and external beam radiotherapy and non established therapies such as minimal invasive techniques and early hormone-therapy (43). RP is considering the gold standard for the treatment of localized PC and in fact it is the most common treatment with approximately 60,000 operations performed annually in the US (44). However, only a small number of elderly patients with early stage PC are treated with RP (45). The reason why advanced age is an unfavorable predictor of the probability of surgical treatment is not known (46). Actually, the fact that elderly individuals have lower life expectancy as well as the belief that elderly patients with localized disease are considering more prone to die with PC than of it, partly explain why PR is the less popular treatment of early PC in elderly patients. Moreover, elderly patients are often being considering fragile enough to receive surgical treatment. Whether and if age increase surgical risk is a controversial issue and for this reason several investigators claim that it is co-morbidity that actually increases the surgical risk and not ageing itself (47). Although, co-morbid illness has demonstrated increasing importance as a prognostic factor, its role is poorly defined. It is generally accepted that co-morbidity limits the generalization of results to older and sicker patients however; the widespread integration of co-morbidity into clinical practice has yet to be

Reported differences in PC specific survival across age strata may be associated with under use of potentially curative local therapy among older men. In fact, PC mortality increases with ageing, peaks at the age of 70-75 and no significant decrease occurs thereafter (table 1). According to the SEER database, younger men (under age 65) with localized prostate cancer had 25-year prostate cancer mortality rates of approximately 19% for Gleason 6 disease, 37% for Gleason 7 disease, and 50% for Gleason 8-10 disease (48). Given that the survival advantage of surgery is most pronounced in men with higher stage disease, it became obvious that elderly PC patients with aggressive disease and life expectancy >10 years are likely to die from progressive prostate cancer (49). Worth mentioning, Bechis et al. studied men in the Cancer of the Prostate Strategic Urologic Research Endeavor (CaPSURE) database with complete risk, treatment, and follow-up information. They found that older patients are more likely to have high-risk prostate cancer at diagnosis and less likely to receive local therapy (30). In confirmation to the above, Dahm et al.showed that risk of death from PC for elderly PC patients treated with PR is significantly lower when compared with

realized.


Table 1. Age and PC specific mortality (US Public service 1989).

that of elderly PC patients treated with watchful waiting (50). Results from other studies showed that surgical therapy can achieve excellent oncologic results in selected elderly patients but they didn't found significant differences in overall survival (51,52). According to the results of the Scandinavian Prostate Cancer Group (SPCG) study, radical prostatectomy is associated with less deaths from prostate cancer (10 vs.15%), less deaths from any other cause (24 vs. 30%) and less metastases (14 vs. 23%) in a median follow-up of 8.2 years. However, benefit in cancer specific survival is limited to patients younger than 65 years.

On the other hand there are several facts supporting WW (deferring intervention until the advent of symptoms) as an ideal treatment of early PC in elderly patients: Epidemiological evidence from autopsy studies show that while a very high proportion of elderly men has histological evidence of the disease, a much smaller proportion actually develop clinically apparent PC (53). Several authors demonstrated that elderly patients with localized PC have a favorable outlook following WW (54,55) and other showed that WW results in similar overall survival when compared with RP (56). In a pooled analysis of 828 case records from six nonrandomized studies, of men treated conservatively for clinically localized prostate cancer, Chodak et al, found an impressive 87% five years disease-specific survival rate (57), however other found that disease specific survival is better in patients who had undergone surgery and some authors argue that WW simply postpone the final treatment (58,59,60,61).

Notably, there are no randomized clinical trials comparing surgery with radiation therapy in elderly PC patients, however an observational study, by Albertsen and colleagues showed that surgery is superior to radiation in localized prostate cancer in terms of prolonging overall and disease-specific survival (62).

The truth is that the preferred management of clinically localized prostate cancer is not known, due in large part to the paucity of randomized controlled trials comparing the effectiveness and harms across primary treatment options. It seems that age itself is the main determinant of treatment selection: according to the Swedish Cancer Register in men with localized tumors expectant treatment was much more commonly used in those aged > or =75 years than in those aged <75 years (63). It is also clear that WW is an adequate

Elderly and Early Prostate Cancer 109

than one uncontrolled comorbidity or severe malnutrition) are 'frail' and should receive adapted treatment; patients in Group 4 (dependent) should receive only symptomatic

Actually, health status is more reliable prognostic factor for survival and treatment related outcomes in oncology than patient age and with this modern approach should be adapted in order to screening senior adults. Age, PSA level, histological grade, and comorbidities should be carefully balanced before making a treatment decision, in elderly men suffering from prostate cancer. Elderly men with limited life expectancy due to other significant lifelimiting medical conditions, such as chronic obstructive pulmonary disease and advanced coronary artery disease, are less likely to benefit from aggressive treatment and are candidate for a palliative approach Therefore, it is reasonable to withhold early detection through PSA screening in these patients thus avoiding the associated risks and impact on quality of life. In selected cases of healthy elderly patients and long life expectancy, PSA screening and curative treatment of undifferentiated prostate cancers could be considered as

[1] Jemal A, Thomas A, Murray T, Thun M. Cancer statistics, 2002. CA Cancer J Clin

Report of the Secretary-General to the 33rd session of the Commission on

challenges for development and modernization: Hong Kong and the Asia-Pacific region in the new millennium. Hong Kong: Chinese University Press; 2002:207-

Patients living with cancer, long-term survivors and cured patients. Epidemiol

Androgen receptor CAG repeat lengths in prostate cancer: correlation with age of

[2] Levy I. Prostate cancer: the epidemiology perspective. Can J Oncol 1994;4 Suppl 1:4-7. [3] United Nations Economic and Social Council.Concise report on world population (2000).

http://www.un.org/documents/ecosoc/cn9/2000/ecn92000-3.pdf. [4] Siegel R, Ward E, et al. Cancer statistics, 2007. CA Cancer J Clin 2007;57:43–66.

[5] Jemal, A; Cheung FM. Ageing population and gender issues. In: Yeung YM, ed. New

[7] AIRTUM Working Group. Italian cancer figures, Report 2010: Cancer prevalence in Italy.

[8] Schutze U. Latent prostatic carcinoma -an autopsy study of men over 50 years of age.

[9] Breslow N, Chan CW, Dhom G, Drur-y RA, Franks LM, Gellei B et al. Latent carcinoma

[10] Hardy DO, Scher HI, Bogenreider T, Sabbatini P, Zhang ZF, Nanus DM, Catterall JF.

Population and Development, 27-31 March 2000 available at:

[6] Crawford, ED. Epidemiology of prostate cancer. Urology 2003;62(6):3–12.

of prostate in seven areas. Int J Cancer 1977;20:680–688.

onset. J Clin Endocrinol Metab. 1996;81:4400-4405

Prev. 2010;34 (5-6 Suppl 2):1-188.

Zentralbl Allg Pathol. 1984;129(4):357-64.

palliative treatment (32).

**4. Conclusions** 

a rational choice.

**5. References** 

23.

2002;52:23-47.

approach for the treatment of early stage PC in patients suffering of live threatening diseases, unfit for radical treatment, however, it remains unclear whether and if treatment can be delayed until absolutely necessary with no detriment to curability in otherwise healthy elderly PC patients. Interestingly, Wong et al found no survival advantage associated with expectant treatment for localized PC in elderly men aged 65 to 80 years (64). Certainly observational data cannot completely adjust for potential selection bias and confounding, however these results clearly shows that specific factors other than tumor stage may contribute to WW failure. Given that PC exhibits a wide range of biologic behaviour, it could be assumed that disease specific survival outcomes in patients with localised PC following WW are associated with Gleason score or baseline PSA level: In the study of Johansson, only the 6% of patients with well differentiated PC, died of PC while mortality rates for intermediate and poorly differentiated cancers were 17% and 56% respectively (54). Soloway and associates reported an 85% treatment-free rate at 5 years on a small cohort of patients diagnosed with 'low-risk' prostate cancer managed by WW (65). Sandblom et al found also a great influence on survival and suggest the grade of malignancy to be taken into account when deciding on therapy (55).

The major risk of watchful waiting is that without treatment, prostate cancer can grow and spread outside the prostate capsule. In fact, even small, slow-growing tumors may become rapidly growing tumors and sometimes prostate cancer that appears to be small and slow growing may be larger and more aggressive than originally thought. Identification of patients who have a low probability of disease progression could be based on strict clinical and pathologic criteria such as Gleason score of 6 or less, a PSA level of 10 ng/ml or less, and stage T1c–T2a disease (66). Again, although patients with these characteristics have a much more favourable natural history and progression rate than those who have a higher Gleason grade or PSA level, in a substantial proportion of men tumours will still progress to advanced, incurable prostate cancer and death (67).

These data suggest that it is of outmost importance to distinguish between patients who are at higher risk and need active therapy and patients who are at low risk for disease progression and support making decisions regarding treatment on the basis of disease risk and life expectancy rather than on chronologic age (68,69).

It therefore became clear that a comprehensive health status assessment is the key in distinguishing between frail and healthy elderly patients and in developing appropriate management approaches for these individuals. The geriatric assessment differs from a standard medical evaluation as it focuses on elderly individuals with complex problems and emphasizes functional status, co-morbidity and quality of life. Most importantly, comprehensive geriatric assessment frequently takes advantage of an interdisciplinary team of providers (urologists, radiation oncologists, medical oncologists and geriatricians).

Recently, the SIOG has developed a proposal of recommendations in this setting based on a systematic bibliographical search focused on screening, diagnostic procedures and treatment options for localised, locally advanced and metastatic prostate cancer in senior adults. Specific aspects of the geriatric approach were emphasised, including evaluation of health status (nutritional, cognitive, thymic, physical and psycho-social) and screening for vulnerability and frailty (70). According to the above elderly PC patients are classified in 4 groups. In Group 1 (no abnormality), patients are 'fit' and should receive the same treatment as younger patients; patients in Group 2 (one impairment in IADL or one uncontrolled comorbidity or at risk of malnutrition) are 'vulnerable' and should receive standard treatment after medical intervention; patients in Group 3 (one impairment in ADL or more than one uncontrolled comorbidity or severe malnutrition) are 'frail' and should receive adapted treatment; patients in Group 4 (dependent) should receive only symptomatic palliative treatment (32).

#### **4. Conclusions**

108 Prostate Cancer – Original Scientific Reports and Case Studies

approach for the treatment of early stage PC in patients suffering of live threatening diseases, unfit for radical treatment, however, it remains unclear whether and if treatment can be delayed until absolutely necessary with no detriment to curability in otherwise healthy elderly PC patients. Interestingly, Wong et al found no survival advantage associated with expectant treatment for localized PC in elderly men aged 65 to 80 years (64). Certainly observational data cannot completely adjust for potential selection bias and confounding, however these results clearly shows that specific factors other than tumor stage may contribute to WW failure. Given that PC exhibits a wide range of biologic behaviour, it could be assumed that disease specific survival outcomes in patients with localised PC following WW are associated with Gleason score or baseline PSA level: In the study of Johansson, only the 6% of patients with well differentiated PC, died of PC while mortality rates for intermediate and poorly differentiated cancers were 17% and 56% respectively (54). Soloway and associates reported an 85% treatment-free rate at 5 years on a small cohort of patients diagnosed with 'low-risk' prostate cancer managed by WW (65). Sandblom et al found also a great influence on survival and suggest the grade of malignancy

The major risk of watchful waiting is that without treatment, prostate cancer can grow and spread outside the prostate capsule. In fact, even small, slow-growing tumors may become rapidly growing tumors and sometimes prostate cancer that appears to be small and slow growing may be larger and more aggressive than originally thought. Identification of patients who have a low probability of disease progression could be based on strict clinical and pathologic criteria such as Gleason score of 6 or less, a PSA level of 10 ng/ml or less, and stage T1c–T2a disease (66). Again, although patients with these characteristics have a much more favourable natural history and progression rate than those who have a higher Gleason grade or PSA level, in a substantial proportion of men tumours will still progress to

These data suggest that it is of outmost importance to distinguish between patients who are at higher risk and need active therapy and patients who are at low risk for disease progression and support making decisions regarding treatment on the basis of disease risk

It therefore became clear that a comprehensive health status assessment is the key in distinguishing between frail and healthy elderly patients and in developing appropriate management approaches for these individuals. The geriatric assessment differs from a standard medical evaluation as it focuses on elderly individuals with complex problems and emphasizes functional status, co-morbidity and quality of life. Most importantly, comprehensive geriatric assessment frequently takes advantage of an interdisciplinary team of providers (urologists, radiation oncologists, medical oncologists and geriatricians).

Recently, the SIOG has developed a proposal of recommendations in this setting based on a systematic bibliographical search focused on screening, diagnostic procedures and treatment options for localised, locally advanced and metastatic prostate cancer in senior adults. Specific aspects of the geriatric approach were emphasised, including evaluation of health status (nutritional, cognitive, thymic, physical and psycho-social) and screening for vulnerability and frailty (70). According to the above elderly PC patients are classified in 4 groups. In Group 1 (no abnormality), patients are 'fit' and should receive the same treatment as younger patients; patients in Group 2 (one impairment in IADL or one uncontrolled comorbidity or at risk of malnutrition) are 'vulnerable' and should receive standard treatment after medical intervention; patients in Group 3 (one impairment in ADL or more

to be taken into account when deciding on therapy (55).

advanced, incurable prostate cancer and death (67).

and life expectancy rather than on chronologic age (68,69).

Actually, health status is more reliable prognostic factor for survival and treatment related outcomes in oncology than patient age and with this modern approach should be adapted in order to screening senior adults. Age, PSA level, histological grade, and comorbidities should be carefully balanced before making a treatment decision, in elderly men suffering from prostate cancer. Elderly men with limited life expectancy due to other significant lifelimiting medical conditions, such as chronic obstructive pulmonary disease and advanced coronary artery disease, are less likely to benefit from aggressive treatment and are candidate for a palliative approach Therefore, it is reasonable to withhold early detection through PSA screening in these patients thus avoiding the associated risks and impact on quality of life. In selected cases of healthy elderly patients and long life expectancy, PSA screening and curative treatment of undifferentiated prostate cancers could be considered as a rational choice.

#### **5. References**


http://www.un.org/documents/ecosoc/cn9/2000/ecn92000-3.pdf.


Elderly and Early Prostate Cancer 111

[27] Brassell SA, Rice KR, Parker PM, et al. Prostate cancer in men 70 years old or older,

[28] Schröder FH, Hugosson J, Roobol MJ, Tammela TL, Ciatto S, Nelen V, Kwiatkowski

[29] Kopec JA, Goel V, Bunting PS, Neuman J, Sayre EC, Warde P, Levers P, Fleshner N.

[30] Bechis SK, Carroll PR, Cooperberg MR. Impact of age at diagnosis on prostate cancer

[31] Advisory Committee on Cancer Prevention. Position paper. Recommendations on

[32] Bennett CL, Greenfield S, Aronow H, Ganz P, Vogelzang NJ, Elashoff RM. Patterns of care related to age of men with prostate cancer. Cancer 1991;67(10):2633-41. [33] Harlan LC, Potosky A, Gilliland FD, Hoffman R, Albertsen PC, Hamilton AS, Eley JW,

[34] Samet, J; Hunt, WC; Key, C, et al. Choice of cancer therapy varies with age of patient.

[35] Lu-Yao, GL; McLerran, D; Wasson, J; Wennberg, JE. An assessment of radical

[36] Sverson, RK; Montie, JE; Porter, AT; Demers, RY. Recent trends in incidence and

[37] Schrag D, Cramer LD, Bach PB, Begg CB. Age and adjuvant chemotherapy use after

[38] Mahoney T, Kuo YH, Topilow A, Davis JM. Stage III colon cancers: why adjuvant

[39] O'Connell JB, Maggard MA, Ko CY. Cancer-directed surgery for localized disease:

[40] Fuchshuber P. Age and Cancer Surgery: Judicious Selection or Discrimination? Ann

[41] Krahn, MD; Bremner, KE; Asaria, J, et al. The ten-year rule revisited: accuracy of

clinicians' estimates of life expectancy in patients with localized prostate cancer.

surgery for stage III colon cancer. J Natl Cancer Inst 2001;93:850 –7.

therapy is not offered to elderly patients. Arch Surg 2000;135:182–5.

decreased utilization in the elderly. Ann Surg Oncol 2004;962–969.

Patient Outcomes Research Team. JAMA. 1993;269:2633–2636.

population based case-control study. J Urol. 2005;174(2):495-9

cancer screening in European Union. Eur J Cancer 2000;36:1473–8

treatment and survival. J Clin Oncol. 2011;29:235-41.

Registry of Sweden. Cancer. 2005;103:943–951.

2011;185:132-7.

2009;360(13):1320-8

2001;93(24):1864-71.

534.

JAMA 1986;255:3385–3390.

Surg Oncol. 2004;11(11):951–952

Urology. 2002;60:258–263.

results from three counties in the population-based National Prostate Cancer

indolent or aggressive: clinicopathological analysis and outcomes. J Urol.

M, Lujan M, Lilja H, Zappa M, Denis LJ, Recker F, Berenguer A, Määttänen L, Bangma CH, Aus G, Villers A, Rebillard X, van der Kwast T, Blijenberg BG, Moss SM, de Koning HJ, Auvinen A; ERSPC Investigators. Screening and prostatecancer mortality in a randomized European study. N Engl J Med.

Screening with prostate specific antigen and metastatic prostate cancer risk: a

Stanford JL, Stephenson RA. Factors associated with initial therapy for clinically localized prostate cancer: prostate cancer outcomes study. J Natl Cancer Inst.

prostatectomy. Time trends, geographic variation, and outcomes. The Prostate

treatment of prostate cancer among elderly men. J Natl Cancer Inst. 1995;87:532–


[11] Fu VX, Dobosy JR, Desotelle JA, Almassi N, Ewald JA, Srinivasan R, Berres M, Svaren J,

[12] Smith DS, Catalona WJ, Herschman JD. Longitudinal screening for prostate cancer with

[13] Woolf HS. Screening for prostate cancer with prostate- specific antigen: an examination

[14] Cooperberg, MR; Lubeck, DP; Mehta, SS; Carroll, PR. Time trends in clinical risk

[15] Epstein JI, Walsh PC, Carmichael M, Brendler CB. Pathologic and clinical findings to

[16] Johansson, JE. Expectant management of early stage prostatic cancer: Swedish

[17] Albertsen PC, Hanley JA, Gleason DF, Barry MJ. Competing risk analysis of men aged

[18] Minino A, Smith BL. Deaths: preliminary data for 2000. Natl Vital Stat Rep. 2001;49:1–40

[19] Horan AH, McGehee M. Mean time to cancer-specific death of apparently clinically

[20] Scales C, Curtis L, Norris R, Schulman K, Albala D, Moul J. Prostate specific antigen testing in men older than 75 years in the United States. J Urol 2006;176(2):511–4. [21] Haggerty J, Tudiver F, Brown JB, Herbert C, Ciampi A, Guibert R. Patients' anxiety and

[22] Hoffman RM, Barry MJ, Stanford JL, Hamilton AS, Hunt WC, Collins MM. Health

[23] Walter LC, Bertenthal D, Lindquist K, Konety BR. PSA screening among elderly men

[24] Bowen DJ, Hannon PA, Harris JR, Martin DP. Prostate cancer screening and informed

[25] D'Ambrosio GG, Campo S, Cancian M, Pecchioli S, Mazzaglia G. Opportunistic

[26] Aus G, Robinson D, Rosell J, Sandblom G, Varenhorst E, for the South-East Region

with limited life expectancies. Nat Clin Pract Urol. 2007;4(10):532-3.

antigen screening and ablative therapy. BJU Int. 2000;85:1063–1066

screening tests. Can Fam Physician. 2005;51(12):1659.

Cancer Outcomes Study. Am J Med. 2006;119(5):418-25.

general practice database. Eur J Cancer Prev. 2010;19:413-6.

prostate-specific antigen. JAMA. 1996;276:1309–1315.

of evidence. N Engl J Med 1995;333:1401–5.

experience. J Urol. 1994;152:1753–1756.

National Statistics. Washton D.C.1983.

cancer. JAMA. 1998;280:975-980.

2011[Epub ahead of print].

Urol. 2003;170(6 Pt 2):S21–S25.

1994;271:368–374.

802.

Weindruch R, Jarrard DF.Aging and cancer-related loss of insulin-like growth factor 2 imprinting in the mouse and human prostate. Cancer Res. 2008;68(16):6797-

stratification for prostate cancer: implications for outcomes (data from CaPSURE). J

predict tumor extent of nonpalpable (stage T1c) prostate cancer. JAMA.

55 to 74 years at diagnosis managed conservatively for clinically localized prostate

No author listed. US Public service, National Statistics Division. "Prostate cancer"

localized prostate cancer: policy implications for threshold ages in prostate-specific

expectations. How they influence family physicians' decisions to order cancer

outcomes in older men with localized prostate cancer: results from the Prostate

decision-making: provider and patient perspectives. Prostate Cancer Prostatic Dis.

prostate-specific antigen screening in Italy: 6 years of monitoring from the Italian

Prostate Cancer Group. Survival in prostate carcinoma–outcome from a prospective, population-based cohort of 8887 men with up to 15 years of follow-up: results from three counties in the population-based National Prostate Cancer Registry of Sweden. Cancer. 2005;103:943–951.


Elderly and Early Prostate Cancer 113

[55] Sandblom G, Dufmats M, Varenhorst E. Long-term survival in a Swedish population-

[56] Iversen P, Madsen PO, Corle DK. Radical prostatectomy versus expectant treatment for

[57] Chodak GW, Thisted RA, Gerber GS, Johansson JE, Adolfsson J, Jones GW, Chisholm

[58] Bill-Axelson A, Holmberg L, Filen F, et al. Radical prostatectomy versus watchful

[59] McLaren DB, Watchful waiting or watchful progression? Prostate specific antigen

[60] Stattin P, Holmberg E, Bratt O, Adolfsson J, Johansson JE, Hugosson J; National Prostate

[61] Makarov MV, Partin AW. Conflicting insights into the role of watchful waiting in the management of adenocarcinoma of the prostate Rev Urol. 2006;8(4):232-234 [62] Albertsen PC, Hanley JA, Penson DF, Fine J. Ten year outcomes following treatment for

[63] Adolfsson J, Garmo H, Varenhorst E, Ahlgren G, Ahlstrand C, Andrén O, Bill-Axelson

[64] Wong YN, Mitra N, Hudes G, Localio R, Schwartz JS, Wan F, Montagnet C, Armstrong

[65] Soloway MS, Soloway CT, Williams S, Ayyathurai R, Kava B, Manoharan M. Active

[66] Epstein JI, Chan DW, Sokoll LJ, Walsh PC, Cox JL, Rittenhouse H, Wolfert R, Carter HB.

[67] Thaxton CS, Loeb S, Roehl KA, Kan D, Catalona WJ. Treatment Outcomes of Radical

early carcinoma of the prostate. Twenty-three year follow-up of a prospective

GD, Moskovitz B, Livne PM, Warner J. Results of conservative management of

waiting in localized prostate cancer: The Scandinavian Prostate Cancer Group-4

doubling times and clinical behavior in patients with early untreated prostate

Cancer Register. Surveillance and deferred treatment for localized prostate cancer. Population based study in the National Prostate Cancer Register of Sweden. J Urol.

clinically localized prostate cancer: a population based study. Program and abstracts of the American Urological Association 2006 Annual Meeting; May 20-25,

A, Bratt O, Damber JE, Hellström K, Hellström M, Holmberg E, Holmberg L, Hugosson J, Johansson JE, Petterson B, Törnblom M, Widmark A, Stattin P. Clinical characteristics and primary treatment of prostate cancer in Sweden between 1996

K. Survival associated with treatment vs observation of localized prostate cancer in

surveillance; a reasonable management alternative for patients with prostate

Nonpalpable stage T1c prostate cancer: prediction of insignificant disease using free/ total prostate specific antigen levels and needle biopsy findings. J Urol

Prostatectomy in Potential Candidates for 3 Published Active Surveillance

based cohort of men with prostate cancer. Urology 2000;56(3):442-7

clinically localized prostate cancer. N Engl J Med. 1994;330(4):242-8

randomized study. Scan J Urol Nephrol Suppl 1995;172:65–72.

randomized trial. J Natl Cancer Inst 2008;100:1144–54.

carcinoma. Cancer. 1998;82:342–348

2006; Atlanta, Georgia. Abstract 652.

and 2005. Scand J Urol Nephrol. 2007;41(6):456-77.

cancer: the Miami experience. BJU Int. 2008;101(2):165-9.

elderly men. JAMA. 2006; 296(22):2683-93.

Protocols. Urology. 2010;75(2):414-8.

2008;180(6):2423-9

1998;160: 2407–2411


[42] Alibhai, SMH; Krahn, MD; Cohen, MM, et al. Is there age bias in the treatment of

[43] Anandadas CN, Clarke NW, Davidson SE, O'Reilly PH, Logue JP, Gilmore L, Swindell

[44] Healthcare Cost and Utilization Project (U.S.). United States Agency for Healthcare

[45] Schwartz KL, Alibhai SM, Tomlinson G, Naglie G, Krahn MD. Continued

[46] Carter HB, Epstein JI, Partin AW. Influence of age and prostate-specific antigen on the

[47] Roberts CB, Albertsen PC, Shao YH, Moore DF, Mehta AR, Stein MN, Lu-Yao GL.

[48] Wong YN, Wan F, Mitra N, et al. Treatment of localized prostate cancer: a survival

[49] Inman BA, Slezak JM, Kwon ED, et al. 25-year outcomes of radical prostatectomy for

[50] Dahm P, Silverstein AD, Weizer AZ, Crisci A, Vieweg J, Paulson DF. When to diagnose

[51] Wilt TJ, Brawer MK, Barry MJ, Jones KM, Kwon Y, Gingrich JR, Aronson WJ, Nsouli I,

[52] Thompson RH, Slezak JM, Webster WS, Lieber MM. Radical prostatectomy for

[53] Pienta K, Esper PS, Risk Factors for Prostate Cancer Ann Int Med 1993; 118(10):793-803 [54] Johansson JE, Holmberg L, Johansson S, Bergström R, Adami HO. Fifteen-year survival

R, Brough RJ, Wemyss-Holden GD, Lau MW, Javle PM, Ramani VA, Wylie JP, Collins GN, Brown S, Cowan RA; North West Uro-oncology Group. Early prostate cancer--which treatment do men prefer and why? BJU Int. 2011 Jun;107(11):1762-8.

Research and Quality. http://www.ahrq.gov/data/hcup/hcupnet.htm. Accessed

undertreatment of older men with localized prostate cancer. Urology. 2003

chance of curable prostate cancer among men with non palpable disease. Urology.

Patterns and correlates of prostate cancer treatment in older men. Am J Med. 2011

analysis using SEER-Medicare data. Program and abstracts of the American Urological Association 2006 Annual Meeting; May 20-25, 2006; Atlanta, Georgia.

the treatment of all stages of non-metastatic prostate cancer. Program and abstracts of the American Urological Association 2006 Annual Meeting; May 20-25, 2006;

and how to treat prostate cancer in the "not too fit" elderly. Crit Rev Oncol

Iyer P, Cartagena R, Snider G, Roehrborn C, Fox S. The Prostate cancer Intervention Versus Observation Trial:VA/NCI/AHRQ Cooperative Studies Program #407 (PIVOT): design and baseline results of a randomized controlled trial comparing radical prostatectomy to watchful waiting for men with clinically localized prostate

octogenarians: how old is too old? Urology. 2006 Nov;68(5):1042-5. Epub 2006

in prostate cancer. A prospective, population-based study in Sweden. JAMA.

localized prostate carcinoma? Cancer. 2004;100:72–81.

doi: 10.1111/j.1464-410X.2010.09833.x. Epub 2010 Nov 17

December 2006.)

Nov;62(5):860-5.

1999;53(1):126-30.

Mar;124(3):235-43

Abstract 658.

Nov 7.

1997;277(6):467-71.

Atlanta, Georgia. Abstract 646.

cancer. Contemp Clin Trials. 2009;30(1):81-7.

Hematol. 2003;48(2):123-31


**8** 

*México, D.F.* 

−) and hydrogen peroxide (H2O2) are

**Tumoral Markers in Prostate Cancer** 

Noemí Cárdenas-Rodríguez and Esaú Floriano-Sánchez *Sección de Posgrado e Investigación, Instituto Politécnico Nacional Laboratorio de Bioquímica y Biología Molecular, Escuela Médico Militar* 

In Mexico, in 70% of cases, the prostate cancer (PCa) is found in advanced stage. PCa currently occupies second place in frequency of cancer in men, surpassed only by skin cancer, and is the second principal cause of death in men after of lung cancer (Hall et al.,

found in a large number of tumors and in high levels they induce cell death, apoptosis,

One of the major sources of ROS is NADPH oxidase (NOX). The NOX are a family of enzymes that are found in various tissues. The NOX receives an electron from NADPH

isozymes increase in association with ROS-production and tumor progression in ovarian and human colon cancer and in DU-145 cells of PCa, respectively (Brar et al., 2003; Lim et

Cells have different antioxidant systems including low molecular weight antioxidant molecules and various antioxidant enzymes. Superoxide dismutase (SOD) catalyses the

(Genkinger et al., 2006). Mn-SOD is the major antioxidant in the mitochondria and is essential to the vitality of mammalian cells. In many types of tumor cells has been found to contain high levels of Mn-SOD, Cu/Zn-SOD or CAT expression compared to their nonmalignant counterpart such as in human tumor cancer cells of esophageal, gastric, ovary, breast, neuroblastoma, osteosarcoma, melanoma, pleura and leukemia (Grigolo et al., 1998; Janssen et al., 2000; Starcevic et al., 2003; Qian et al., 2005, López Laur et al., 2008).

On the other hand, iNOS or NOS-2 is an inducible isoform of nitric oxide synthases (NOS). All isoforms of NOS catalyze the reaction of L-arginine, NADPH and oxygen to nitric oxide (NO), L-citrulline and NADP. NO is a lipophilic physiological messenger wich regulate a variety of cellular responses and may exert its cellular action by cGMP-dependent as well as by cGMP-independent pathways (Stamler, 1994). The expression of iNOS has been found to be increased in a variety of human cancers such as colon, stomach, brain and breasts cancers (Alderton et al., 2001; Church & Fulton, 2006) by multiple mechanisms that control their

(Bánfi et al., 2001). Xia et al, Lim et al. and Brar et al. found that some NOX

into H2O2 that can be transformed into H2O and O2 by catalase (CAT)

**1. Introduction** 

Reactive oxygen species (ROS) such as superoxide (O2

senescence and angiogenesis (Ushio-Fukai & Nakamura, 2008).

However, the role of these enzymes in carcinogenesis remains unclear.

activity (Stamler, 1994; Friebe & Koesling, 2003).

2005).

generating O2

al., 2005; Xia et al., 2007).

dismutation of O2

*Laboratorio de Neuroquímica, Instituto Nacional de Pediatría* 


### **Tumoral Markers in Prostate Cancer**

Noemí Cárdenas-Rodríguez and Esaú Floriano-Sánchez

*Sección de Posgrado e Investigación, Instituto Politécnico Nacional Laboratorio de Bioquímica y Biología Molecular, Escuela Médico Militar Laboratorio de Neuroquímica, Instituto Nacional de Pediatría México, D.F.* 

#### **1. Introduction**

114 Prostate Cancer – Original Scientific Reports and Case Studies

[68] Klotz L. Active surveillance with selective delayed intervention: using natural history

[69] Berglund A, Garmo H, Tishelman C, Holmberg L, Stattin P, Lambe M. Comorbidity,

[70] Droz JP, Balducci L, Bolla M, Emberton M, Fitzpatrick JM, Joniau S, Kattan MW,

cancer in senior adults. Crit Rev Oncol Hematol. 2010 Jan;73(1):68-91.

PCBaSe Sweden. Urol. 2011;185(3):833-9

50.

to guide treatment in good risk prostate cancer. J Urol. 2004 Nov;172(5 Pt 2):S48-

treatment and mortality: a population based cohort study of prostate cancer in

Monfardini S, Moul JW, Naeim A, van Poppel H, Saad F, Sternberg CN. Background for the proposal of SIOG guidelines for the management of prostate

> In Mexico, in 70% of cases, the prostate cancer (PCa) is found in advanced stage. PCa currently occupies second place in frequency of cancer in men, surpassed only by skin cancer, and is the second principal cause of death in men after of lung cancer (Hall et al., 2005).

> Reactive oxygen species (ROS) such as superoxide (O2 −) and hydrogen peroxide (H2O2) are found in a large number of tumors and in high levels they induce cell death, apoptosis, senescence and angiogenesis (Ushio-Fukai & Nakamura, 2008).

> One of the major sources of ROS is NADPH oxidase (NOX). The NOX are a family of enzymes that are found in various tissues. The NOX receives an electron from NADPH generating O2 (Bánfi et al., 2001). Xia et al, Lim et al. and Brar et al. found that some NOX isozymes increase in association with ROS-production and tumor progression in ovarian and human colon cancer and in DU-145 cells of PCa, respectively (Brar et al., 2003; Lim et al., 2005; Xia et al., 2007).

> Cells have different antioxidant systems including low molecular weight antioxidant molecules and various antioxidant enzymes. Superoxide dismutase (SOD) catalyses the dismutation of O2 into H2O2 that can be transformed into H2O and O2 by catalase (CAT) (Genkinger et al., 2006). Mn-SOD is the major antioxidant in the mitochondria and is essential to the vitality of mammalian cells. In many types of tumor cells has been found to contain high levels of Mn-SOD, Cu/Zn-SOD or CAT expression compared to their nonmalignant counterpart such as in human tumor cancer cells of esophageal, gastric, ovary, breast, neuroblastoma, osteosarcoma, melanoma, pleura and leukemia (Grigolo et al., 1998; Janssen et al., 2000; Starcevic et al., 2003; Qian et al., 2005, López Laur et al., 2008). However, the role of these enzymes in carcinogenesis remains unclear.

> On the other hand, iNOS or NOS-2 is an inducible isoform of nitric oxide synthases (NOS). All isoforms of NOS catalyze the reaction of L-arginine, NADPH and oxygen to nitric oxide (NO), L-citrulline and NADP. NO is a lipophilic physiological messenger wich regulate a variety of cellular responses and may exert its cellular action by cGMP-dependent as well as by cGMP-independent pathways (Stamler, 1994). The expression of iNOS has been found to be increased in a variety of human cancers such as colon, stomach, brain and breasts cancers (Alderton et al., 2001; Church & Fulton, 2006) by multiple mechanisms that control their activity (Stamler, 1994; Friebe & Koesling, 2003).

Tumoral Markers in Prostate Cancer 117

biotinylated peroxidase complex (ABC-kit Vectastain, Vector Laboratories, Burlingame, CA) and diaminobenzidine (Vector Laboratories, Burlingame, CA). After of intensive washing in PBS, slides were counterstained with hematoxylin. Sections were dehydrated in graded alcohols, treated with xylene and subsequently mounted. All specimens were examined by light microscopy (Axiovert 200 M, Carl Zeiss, Germany), photographs were taken with a digital camera (Axiocam HRC, Carl Zeiss, Germany). The number of positive cells (brown) was determined with a computerized image analyzer KS-300 3.0 (Carl Zeiss, Germany). The percentage of damaged area with histopathological alterations was obtained (400x magnification). Five random fields were studied (total area 1,584,000 2). The results were

Findings were expressed as the mean ± SD. The statistical significance of the protein expression levels of p22 *phox* subunit of NOX, Mn-SOD, Cu/Zn-SOD, CAT, iNOS and COX-2 between PCa and BPH groups glands or stroma, was determined using the software Prism version 3.32 (GraphPad Prism 4.0 Software, San Diego, CA, USA) with "student t-test". It

The results obtained in PCa and BPH groups are summarized in Table 1. NOX, Mn-SOD, Cu/Zn-SOD and CAT protein immunohistochemistry were significantly higher (1.76, 1.7, 1.78 and 5.88 fold, respectively) in stroma and were significantly higher (3.74, 1.69, 4.76 and

Moreover, NOX, Mn-SOD and CAT protein expressions were significantly higher in gland than in stroma, while as Cu/Zn-SOD protein expression was significantly higher in stroma than in gland in patients with BPH. NOX and Mn-SOD protein expression were significantly

However, iNOS and COX-2 protein expressions were significantly higher in stroma and

Parameters BPH (n=32) Gland PCa (n=30) Gland Stroma Stroma NOX 4.8 ± 1.9 6.7 ± 2a 8.45 ± 1.7c 25.08 ± 3.5d,b Mn-SOD 11.97 ± 1.6 14.73 ± 1.4c 20.45 ± 2.1c 24.83 ± 1.7d,b Cu/Zn-SOD 30.3 ± 6.6d 11.1 ± 1.9 54.1 ± 14.6c 52.8 ± 8.8d CAT 9.8 ± 1.5 37.9 ± 4.5c 57.6 ± 15.5c 60.1 ± 4.5d

COX-2 12.1 ± 1.3g 14.8 ± 2.1h 7.4 ± 0.9 5.12± 0.7

Table 1. Mean ± SD NOX, Mn-SOD, Cu/Zn-SOD, CAT, iNOS and COX-2 protein

aP=0.0002 vs stroma BPH; bP<0.0001 vs stroma PCa; cP<0.0001 vs stroma BPH; dP<0.0001 vs gland BPH;

P=0.0016 vs gland PCa; gP=0.0314 vs stroma PCa; hP=0.0072 vs gland PCa

15.9 ± 7.1 16.3 ± 4.6

1.59 fold ,respectively) in gland of patients with PCa than that in patients with BPH.

was considered a p <0.05 as statistical difference between groups.

gland of BPH (1.47 and 2.9 fold, respectively) in comparison with PCa.

higher in gland than in stroma in patients with PCa.

iNOS 23.3 ± 8.8e 24.6 ± 6.3f

expressions (%) in PCa and BPH group.

eP=0.0072 vs stroma PCa; f

expressed as a percentage.

**2.2 Statistics** 

**3. Results** 

Ciclooxygenase-1 and 2 (COX-1/2) catalyze the initial step in the formation of prostaglandins (Smith & Langenbach, 2001). Very recently their role in carcinogenesis has become more evident. They influence apoptosis, angiogenesis, and invasion, and play a key role in the production of carcinogens. Usually, a high level of COX expression is found in cancer cells (Dannenberg & Zakim, 1999). The role of COX-2 in carcinogenesis has been recently described. Multiple lines of evidence confirm that selective COX-2 inhibitors reduce prostaglandin production and the risk of colorectal, skin and other neoplasias (Sonoshita et al., 2001). COX-2 is related to the formation of carcinogens, tumor promotion and inhibition of apoptosis, angiogenesis and the metastatic process (Ebehart et al.,1994; Uefuji et al., 2000). However, the interactions and links between lipid metabolism and cancer progression remain to be elucidated.

Therefore, in the present study, we decided to evaluate and compare, for the first time, the pattern protein expression of p22 *phox* subunit of NOX*,* Mn-SOD, Cu/Zn-SOD, CAT, iNOS and COX-2 protein expression in patients with PCa and with BPH.

#### **2. Patients and methods**

We obtained 62 samples of prostate tissue through of various surgical procedures (transurethral resection and biopsy transrectal). Approval was obtained from the local research and ethics committee for use of tissue. Of these samples, 30 patients (48.4%) had a diagnosis of PCa, while as 32 patients (51.6%) had a diagnosis of BPH (Department of Medical Urology, Hospital Central Militar, Mexico). The sample collection was conducted from January 2006 to December 2009 and was considered inclusion, exclusion and elimination criteria.

In the PCa group the average age was of 65.3 years and the concentration of preoperative PSA was of 8.6 ng/mL. In this group, the patients were classified according to Gleason scale. The score was of 4 in 1 case (3.3%), 6 in 19 cases (63.3%), 7 in 9 cases (30%) and 8 in 1 case (3.3%). None of the patients had undergone chemotherapy or radiotherapy before surgery.

In the BPH group the average age was of 66.5 years and the concentration of preoperative PSA was of 8.7 ng/mL.

Tissues obtained (500 mg) were stored at -83°C (Revco® Legaci ULT2186 3-35 Dupont SVVA Refrigerants) until further processing.

#### **2.1 Immunohistochemistry**

For light microscopy, tissue samples of PCa and BPH were fixed by immersion in formalin (pH 7.4) and embedded in paraffin. Serial cuts of 3 mm of thickness were mounted on poli-L-lisina coated slides (Sigma, St Louis, MO). Sections were initially deparaffinized by washing in xylene and decreasing ethanol concentrations and boiled in Declere (Cell Marque, Hot Springs, AR) to unmask antigen sites. Slides were washed in phosphate buffer saline (PBS). Endogenous peroxidase activity was blocked by exposing slides to 0.6% H2O2 in PBS for 30 min.

After of washing in PBS, nonspecific binding was avoided by incubation with 5% blocking solution (5% normal goat serum in PBS) for 20 min. Sections were incubated overnight (16 h) with primary anti-p22 *phox* subunit NOX, anti-Mn-SOD, anti-Cu/Zn-SOD, anti-CAT, antiiNOS and anti-COX-2 antibody (1:100 for each one). Following removal of the antibodies and repetitive rinsing with PBS, slides were incubated with a biotinylated goat anti-IgG secondary antibodies (1:500 fur each one) (Jackson ImmunoReseach, West Grove, PA). Immunocytochemical identification of positive cells was performed by the use of an avidinbiotinylated peroxidase complex (ABC-kit Vectastain, Vector Laboratories, Burlingame, CA) and diaminobenzidine (Vector Laboratories, Burlingame, CA). After of intensive washing in PBS, slides were counterstained with hematoxylin. Sections were dehydrated in graded alcohols, treated with xylene and subsequently mounted. All specimens were examined by light microscopy (Axiovert 200 M, Carl Zeiss, Germany), photographs were taken with a digital camera (Axiocam HRC, Carl Zeiss, Germany). The number of positive cells (brown) was determined with a computerized image analyzer KS-300 3.0 (Carl Zeiss, Germany). The percentage of damaged area with histopathological alterations was obtained (400x magnification). Five random fields were studied (total area 1,584,000 2). The results were expressed as a percentage.

#### **2.2 Statistics**

116 Prostate Cancer – Original Scientific Reports and Case Studies

Ciclooxygenase-1 and 2 (COX-1/2) catalyze the initial step in the formation of prostaglandins (Smith & Langenbach, 2001). Very recently their role in carcinogenesis has become more evident. They influence apoptosis, angiogenesis, and invasion, and play a key role in the production of carcinogens. Usually, a high level of COX expression is found in cancer cells (Dannenberg & Zakim, 1999). The role of COX-2 in carcinogenesis has been recently described. Multiple lines of evidence confirm that selective COX-2 inhibitors reduce prostaglandin production and the risk of colorectal, skin and other neoplasias (Sonoshita et al., 2001). COX-2 is related to the formation of carcinogens, tumor promotion and inhibition of apoptosis, angiogenesis and the metastatic process (Ebehart et al.,1994; Uefuji et al., 2000). However, the interactions and links between lipid metabolism and cancer progression

Therefore, in the present study, we decided to evaluate and compare, for the first time, the pattern protein expression of p22 *phox* subunit of NOX*,* Mn-SOD, Cu/Zn-SOD, CAT, iNOS

We obtained 62 samples of prostate tissue through of various surgical procedures (transurethral resection and biopsy transrectal). Approval was obtained from the local research and ethics committee for use of tissue. Of these samples, 30 patients (48.4%) had a diagnosis of PCa, while as 32 patients (51.6%) had a diagnosis of BPH (Department of Medical Urology, Hospital Central Militar, Mexico). The sample collection was conducted from January 2006 to

In the PCa group the average age was of 65.3 years and the concentration of preoperative PSA was of 8.6 ng/mL. In this group, the patients were classified according to Gleason scale. The score was of 4 in 1 case (3.3%), 6 in 19 cases (63.3%), 7 in 9 cases (30%) and 8 in 1 case (3.3%). None of the patients had undergone chemotherapy or radiotherapy before surgery. In the BPH group the average age was of 66.5 years and the concentration of preoperative

Tissues obtained (500 mg) were stored at -83°C (Revco® Legaci ULT2186 3-35 Dupont SVVA

For light microscopy, tissue samples of PCa and BPH were fixed by immersion in formalin (pH 7.4) and embedded in paraffin. Serial cuts of 3 mm of thickness were mounted on poli-L-lisina coated slides (Sigma, St Louis, MO). Sections were initially deparaffinized by washing in xylene and decreasing ethanol concentrations and boiled in Declere (Cell Marque, Hot Springs, AR) to unmask antigen sites. Slides were washed in phosphate buffer saline (PBS). Endogenous peroxidase activity was blocked by exposing slides to 0.6% H2O2

After of washing in PBS, nonspecific binding was avoided by incubation with 5% blocking solution (5% normal goat serum in PBS) for 20 min. Sections were incubated overnight (16 h) with primary anti-p22 *phox* subunit NOX, anti-Mn-SOD, anti-Cu/Zn-SOD, anti-CAT, antiiNOS and anti-COX-2 antibody (1:100 for each one). Following removal of the antibodies and repetitive rinsing with PBS, slides were incubated with a biotinylated goat anti-IgG secondary antibodies (1:500 fur each one) (Jackson ImmunoReseach, West Grove, PA). Immunocytochemical identification of positive cells was performed by the use of an avidin-

December 2009 and was considered inclusion, exclusion and elimination criteria.

and COX-2 protein expression in patients with PCa and with BPH.

remain to be elucidated.

**2. Patients and methods** 

PSA was of 8.7 ng/mL.

in PBS for 30 min.

**2.1 Immunohistochemistry** 

Refrigerants) until further processing.

Findings were expressed as the mean ± SD. The statistical significance of the protein expression levels of p22 *phox* subunit of NOX, Mn-SOD, Cu/Zn-SOD, CAT, iNOS and COX-2 between PCa and BPH groups glands or stroma, was determined using the software Prism version 3.32 (GraphPad Prism 4.0 Software, San Diego, CA, USA) with "student t-test". It was considered a p <0.05 as statistical difference between groups.

#### **3. Results**

The results obtained in PCa and BPH groups are summarized in Table 1. NOX, Mn-SOD, Cu/Zn-SOD and CAT protein immunohistochemistry were significantly higher (1.76, 1.7, 1.78 and 5.88 fold, respectively) in stroma and were significantly higher (3.74, 1.69, 4.76 and 1.59 fold ,respectively) in gland of patients with PCa than that in patients with BPH.

Moreover, NOX, Mn-SOD and CAT protein expressions were significantly higher in gland than in stroma, while as Cu/Zn-SOD protein expression was significantly higher in stroma than in gland in patients with BPH. NOX and Mn-SOD protein expression were significantly higher in gland than in stroma in patients with PCa.



aP=0.0002 vs stroma BPH; bP<0.0001 vs stroma PCa; cP<0.0001 vs stroma BPH; dP<0.0001 vs gland BPH; eP=0.0072 vs stroma PCa; f P=0.0016 vs gland PCa; gP=0.0314 vs stroma PCa; hP=0.0072 vs gland PCa

Table 1. Mean ± SD NOX, Mn-SOD, Cu/Zn-SOD, CAT, iNOS and COX-2 protein expressions (%) in PCa and BPH group.

Tumoral Markers in Prostate Cancer 119

Fig. 3. Immunohistochemical determination of iNOS and COX-2 in BPH and PCa. (A) y (B) gland of BPH of iNOS and COX-2. (C) y (D) gland of PCa iNOS and COX-2. In both groups

Recently, a new hypothesis has been proposed for prostate carcinogenesis. It suggested that exposure to environmental factors such as infectious agents and dietary carcinogens, and hormonal imbalances lead to injury of the prostate and to the development of chronic inflammation and regenerative 'risk factor' lesions, referred to as proliferative inflammatory atrophy (PIA). PCa is associated with oxidative stress, which stimulates the production of reactive oxidative species (ROS) and reactive nitrogen species. Oxidative stress derived from endogenous and exogenous sources are associated with DNA damage that occurs with

The results obtained, for the first time, in this study showed an increased in the expression of p22 *phox* subunit of NOX, Mn and Cu/Zn-SOD and CAT in stroma and gland of PCa. In previous studies concluded that NOX has a role as a signaling mechanism that regulates the cell growth and apoptosis in PCa (Vignais, 2002). The exact signaling pathways of NOX

Angiotensin II stimulates the activity of NOX in vascular smooth muscle via protein kinase and NF-B in the airways and in melanomas (Arnold et al., 2001). Arbiser et al. demonstrated that NOX-1-induced vascular endothelial growth factor (VEGF) and VEGF receptor expression promoting the angiogenesis and rapid expansion of the tumors (Arbiser et al., 2002). Babior BM found that the high levels of ROS are produced spontaneously in PCa and in ovarian cancer. This high production of reactive species was inhibited using an inhibitor of NOX, the diphenyl iodonium (DPI) and the inhibitor of mitochondrial electron

was determined % area marked by field (400x) and was analized the values with

significative increase in gland PCa immunoreactivity.

aging and plays a role in carcinogenesis (Klein et al., 2006).

are uncertain and may be tissue specific.

**4. Discussion** 

Fig. 1. Immunohistochemical determination of p22 phox subunit of NOX and Mn-SOD in BPH and PCa. (A) y (B) gland of BPH of NOX and Mn-SOD. (C) y (D) gland of PCa of NOX and Mn-SOD. In both groups was determined % area marked by field (400x) and was analized the values with significative increase in gland PCa immunoreactivity.

Fig. 2. Immunohistochemical determination of Cu/Zn-SOD and CAT in BPH and PCa. (A) y (B) gland of BPH of Cu/Zn-SOD and CAT. (C) y (D) gland of PCa of Cu/Zn-SOD and CAT. In both groups was determined % area marked by field (400x) and was analized the values with significative increase in gland PCa immunoreactivity.

Fig. 1. Immunohistochemical determination of p22 phox subunit of NOX and Mn-SOD in BPH and PCa. (A) y (B) gland of BPH of NOX and Mn-SOD. (C) y (D) gland of PCa of NOX and Mn-SOD. In both groups was determined % area marked by field (400x) and was analized the values with significative increase in gland PCa immunoreactivity.

Fig. 2. Immunohistochemical determination of Cu/Zn-SOD and CAT in BPH and PCa. (A) y (B) gland of BPH of Cu/Zn-SOD and CAT. (C) y (D) gland of PCa of Cu/Zn-SOD and CAT. In both groups was determined % area marked by field (400x) and was analized the values

with significative increase in gland PCa immunoreactivity.

Fig. 3. Immunohistochemical determination of iNOS and COX-2 in BPH and PCa. (A) y (B) gland of BPH of iNOS and COX-2. (C) y (D) gland of PCa iNOS and COX-2. In both groups was determined % area marked by field (400x) and was analized the values with significative increase in gland PCa immunoreactivity.

#### **4. Discussion**

Recently, a new hypothesis has been proposed for prostate carcinogenesis. It suggested that exposure to environmental factors such as infectious agents and dietary carcinogens, and hormonal imbalances lead to injury of the prostate and to the development of chronic inflammation and regenerative 'risk factor' lesions, referred to as proliferative inflammatory atrophy (PIA). PCa is associated with oxidative stress, which stimulates the production of reactive oxidative species (ROS) and reactive nitrogen species. Oxidative stress derived from endogenous and exogenous sources are associated with DNA damage that occurs with aging and plays a role in carcinogenesis (Klein et al., 2006).

The results obtained, for the first time, in this study showed an increased in the expression of p22 *phox* subunit of NOX, Mn and Cu/Zn-SOD and CAT in stroma and gland of PCa.

In previous studies concluded that NOX has a role as a signaling mechanism that regulates the cell growth and apoptosis in PCa (Vignais, 2002). The exact signaling pathways of NOX are uncertain and may be tissue specific.

Angiotensin II stimulates the activity of NOX in vascular smooth muscle via protein kinase and NF-B in the airways and in melanomas (Arnold et al., 2001). Arbiser et al. demonstrated that NOX-1-induced vascular endothelial growth factor (VEGF) and VEGF receptor expression promoting the angiogenesis and rapid expansion of the tumors (Arbiser et al., 2002). Babior BM found that the high levels of ROS are produced spontaneously in PCa and in ovarian cancer. This high production of reactive species was inhibited using an inhibitor of NOX, the diphenyl iodonium (DPI) and the inhibitor of mitochondrial electron

Tumoral Markers in Prostate Cancer 121

NO is synthesized by three differentially gene-encoded NOS in mammals: neuronal NOS (nNOS or NOS-1), inducible NOS (iNOS or NOS-2) and endothelial NOS (eNOS or NOS-3). All three isoforms present similar structures and catalytic modes. The expression of NOS-2 is induced by inflammatory stimuli while NOS-1 and NOS-3 are more or less constitutively expressed. The active form of NOS-1 and -3 requires two NOS monomers associated with two Ca2+-binding protein calmodulin and cofactors such as (6R)-5,6,7,8-tetrahydrobiopterin (BH4), FAD, FMN and haem group and catalyze the reaction of L-arginine, NADPH and oxygen to NO, L-citrulline and NADP (Alderton et al., 2001; Stuehr et al., 2004). NOS isoforms are differentially regulated at transcriptional, translational and post-translational levels. The intracellular localization is relevant for NOS activity. Evidence indicates that NOS are present in plasma membrane, Golgi, cytosol, nucleus and mitochondria (Oess et al., 2006; Iwakiri et al., 2006). The expression of iNOS can be transcriptionally regulated by factors such as cytokines (e.g. interferon-γ (IFN- γ), interleukin-1β (IL-1 β) and tumour necrosis factor-α (TNF-α), bacterial endotoxin (LPS) and oxidative stress (e.g. under

An initial study on iNOS expression in human breast cancer suggested that iNOS activity was higher in less differentiated tumours in a panel of 15 invasive breast carcinomas (Thomsen et al., 1995). Reveneau et al reported NOS activity in 27 of 40 tumours studied (Reveneau et al., 1999). Vakkala et al showed that carcinomas with both iNOS positive tumour and stromal cells had a higher apoptotic index and a higher calculated microvessel density index (Vakkala et al., 2000). Loibl et al further demonstrated that while none of the benign lesions were positive for iNOS, 67% *in situ* carcinomas and 61% invasive lesions

In addition to breast cancer, iNOS has also been shown to be markedly expressed in approximately 60% of human adenomas and in 20-25% of colon carcinomas, while expression was either low or absent in the surrounding normal tissues (Ambs et al., 1998a). In human ovarian cancer, iNOS activity has been localized in tumour cells and not found in normal tissue (Thomsen et al., 1995). Other tumours that have demonstrated iNOS gene expression are brain, head and neck, esophagus, lung, prostate, bladder, pancreatic, and Kaposi's sarcoma (Cobbs et al., 1995; Rosbe et al., 1995; Wilson et al., 1998; Ambs et al.,

In this study, we found that exist strong expression of iNOS in stroma and gland of HPB, in comparison with PCa. It Have been demonstrated that NO- mediated up-regulation of VEGF. In the results is possible that NO generated by iNOS in stroma may promote early new blood vessel formation by up-regulating VEGF and enhance ability of the tumour to grow and increases its invasiveness ability in gland. (Ambs et al., 1998a). Moreover, the accumulation of p53 in gland can result in down-regulation of iNOS expression by inhibition of iNOS promoter activity (Ambs et al., 1998b). On the other hand, the generation of chronic injury and irritation initiate the inflammatory response of stroma to gland (NOS-1). A subsequent respiratory burst an increase uptake of oxygen that leads to the release of

surrounding cells and drive carcinogenesis by altering targets and pathways that are crucial

COX-1 and COX-2 regulate a key step in prostanoid (i.e., tromboxanes and prostaglandins) synthesis. Prostaglandins regulate various pathophysiological processes such as inflammatory reaction, gastrointestinal cytoprotection and ulceration (Smith & Langenbach,

to normal prostate homeostasis (Coussens & Werb, 2002; Fukumura et al., 2006).

, N2O3, NO2 and NO3) from leucocytes can damage

1998a; Klotz et al., 1998; Hajri et al., 1998; Weninger et al., 1998; Swane et al., 1999).

conditions encountered during hypoxia)(Xu & Liu, 1998).

showed iNOS tumour cell staining (Loibl et al., 2002).

reactive oxygen species (NO, ONOO

chain, rotenone (Babior, 1999). This suggest that NOX could promote angiogenesis in the early stages of PCa.

By controlling O2 ─/H2O2 levels, SOD appears to be a critical enzyme in cancer progression. Bravard et al and St Clair et al suggested to the Mn-SOD as a potential tumoral suppressor that might also be involved in cellular differentiation (Zhao et al., 2001).

We suggested that any mutation or epigenetic changes in Mn-SOD gene are the cause of the high level found in the Mn-SOD expression in PCa in the mitochondria. This could have potential effects on survival and proliferation of tumor cells, a fact which has been found in other tumors with aggressive behavior and with a poor prognosis for the patient.

MnSOD polymorphisms have been investigated in several types of malignancies, such as lung, breast and skin cancer (Liu et al., 2004; Han et al., 2007; Bewick et al., 2008). There are at least two functional validated single nucleotide polymorphisms in Mn-SOD. One of these variants is a change in the amino acid codon 9 from valine (GTT) to alanine (GCT) and another is a change in the amino acid codon 16 from valine (GTT) to alanine (GCT) (Tugcu et al., 2007). These changes alter the secondary structure of the protein, affect the transport of the enzyme into mitochondria and reduce the enzymatic activity of Mn-SOD, leaving the cell vulnerable to oxidative damage.

Our results suggest that Mn-SOD probably plays an important role in resistance to treatment of various tumors or in the evolution of invasive tumors.

Brown et al demonstrated an essential role of O2 in the posttranslational activation of Cu/Zn-SOD and in the ratio of active to inactive Cu/Zn-SOD, which may be relevant to various diseases, including cancer (Brown et al., 2004). Therefore, O2 production by NOX could be induce protein over-expression of Cu/Zn-SOD in PCa.

CAT plays an integral role in the primary defense against oxidative stress by converting H2O2 into H2O and O2. Genetic polymorphisms of CAT can change expression levels of the protein. A −262C → T polymorphism in the promoter region of the CAT gene is associated with risk of several conditions related to oxidative stress. A transcription factor binding site search indicates that the -262 C allele is located in close proximity to several binding sites for transcription factors and could potentially influence rates of transcription. Forsberg et al. previously showed that the T allele was associated with greater CAT protein levels in some tissues than the C allele (Forsberg et al., 2001). However, different regulatory mechanism of CAT in PCa should be explained.

Our results showed different expressions in NOX, Mn-SOD, Cu/Zn-SOD and CAT in stroma and gland in PCa and BPH groups. The increase of NOX and Mn-SOD expression in gland of PCa and BPH group may have been due to excessive O2 and H2O2 production that stimulate migration, invasion and angiogenesis of the tumor cells in response to the intracellular changes in ROS levels in this prostate component. The differences in the architecture of the prostate are most likely related to changes in the tumor invasion process

Furthermore our results suggest that exist alterations in the prooxidative-antioxidative balance in PCa, this imbalance is known to alter cellular redox processes, growth, and proliferation and cell cycles, since it is known that certain free radicals mediate the activation of cellular transduction, of transcription factors such as Fos, Jun and nuclear factor kB and an increase in mitochondrial activity in the cells. Moreover, transcription factors such as Rac1, Ref-1 and p53 regulated by ROS are involved in angiogenesis (Ushio-Fukai & Nakayama, 2008).

chain, rotenone (Babior, 1999). This suggest that NOX could promote angiogenesis in the

Bravard et al and St Clair et al suggested to the Mn-SOD as a potential tumoral suppressor

We suggested that any mutation or epigenetic changes in Mn-SOD gene are the cause of the high level found in the Mn-SOD expression in PCa in the mitochondria. This could have potential effects on survival and proliferation of tumor cells, a fact which has been found in

MnSOD polymorphisms have been investigated in several types of malignancies, such as lung, breast and skin cancer (Liu et al., 2004; Han et al., 2007; Bewick et al., 2008). There are at least two functional validated single nucleotide polymorphisms in Mn-SOD. One of these variants is a change in the amino acid codon 9 from valine (GTT) to alanine (GCT) and another is a change in the amino acid codon 16 from valine (GTT) to alanine (GCT) (Tugcu et al., 2007). These changes alter the secondary structure of the protein, affect the transport of the enzyme into mitochondria and reduce the enzymatic activity of Mn-SOD, leaving the

Our results suggest that Mn-SOD probably plays an important role in resistance to

Cu/Zn-SOD and in the ratio of active to inactive Cu/Zn-SOD, which may be relevant to

CAT plays an integral role in the primary defense against oxidative stress by converting H2O2 into H2O and O2. Genetic polymorphisms of CAT can change expression levels of the protein. A −262C → T polymorphism in the promoter region of the CAT gene is associated with risk of several conditions related to oxidative stress. A transcription factor binding site search indicates that the -262 C allele is located in close proximity to several binding sites for transcription factors and could potentially influence rates of transcription. Forsberg et al. previously showed that the T allele was associated with greater CAT protein levels in some tissues than the C allele (Forsberg et al., 2001). However, different regulatory mechanism of

Our results showed different expressions in NOX, Mn-SOD, Cu/Zn-SOD and CAT in stroma and gland in PCa and BPH groups. The increase of NOX and Mn-SOD expression

production that stimulate migration, invasion and angiogenesis of the tumor cells in response to the intracellular changes in ROS levels in this prostate component. The differences in the architecture of the prostate are most likely related to changes in the

Furthermore our results suggest that exist alterations in the prooxidative-antioxidative balance in PCa, this imbalance is known to alter cellular redox processes, growth, and proliferation and cell cycles, since it is known that certain free radicals mediate the activation of cellular transduction, of transcription factors such as Fos, Jun and nuclear factor kB and an increase in mitochondrial activity in the cells. Moreover, transcription factors such as Rac1, Ref-1 and p53 regulated by ROS are involved in angiogenesis (Ushio-

in gland of PCa and BPH group may have been due to excessive O2

in the posttranslational activation of

production by NOX

and H2O2

that might also be involved in cellular differentiation (Zhao et al., 2001).

treatment of various tumors or in the evolution of invasive tumors.

various diseases, including cancer (Brown et al., 2004). Therefore, O2

could be induce protein over-expression of Cu/Zn-SOD in PCa.

Brown et al demonstrated an essential role of O2

other tumors with aggressive behavior and with a poor prognosis for the patient.

─/H2O2 levels, SOD appears to be a critical enzyme in cancer progression.

early stages of PCa. By controlling O2

cell vulnerable to oxidative damage.

CAT in PCa should be explained.

tumor invasion process

Fukai & Nakayama, 2008).

NO is synthesized by three differentially gene-encoded NOS in mammals: neuronal NOS (nNOS or NOS-1), inducible NOS (iNOS or NOS-2) and endothelial NOS (eNOS or NOS-3). All three isoforms present similar structures and catalytic modes. The expression of NOS-2 is induced by inflammatory stimuli while NOS-1 and NOS-3 are more or less constitutively expressed. The active form of NOS-1 and -3 requires two NOS monomers associated with two Ca2+-binding protein calmodulin and cofactors such as (6R)-5,6,7,8-tetrahydrobiopterin (BH4), FAD, FMN and haem group and catalyze the reaction of L-arginine, NADPH and oxygen to NO, L-citrulline and NADP (Alderton et al., 2001; Stuehr et al., 2004). NOS isoforms are differentially regulated at transcriptional, translational and post-translational levels. The intracellular localization is relevant for NOS activity. Evidence indicates that NOS are present in plasma membrane, Golgi, cytosol, nucleus and mitochondria (Oess et al., 2006; Iwakiri et al., 2006). The expression of iNOS can be transcriptionally regulated by factors such as cytokines (e.g. interferon-γ (IFN- γ), interleukin-1β (IL-1 β) and tumour necrosis factor-α (TNF-α), bacterial endotoxin (LPS) and oxidative stress (e.g. under conditions encountered during hypoxia)(Xu & Liu, 1998).

An initial study on iNOS expression in human breast cancer suggested that iNOS activity was higher in less differentiated tumours in a panel of 15 invasive breast carcinomas (Thomsen et al., 1995). Reveneau et al reported NOS activity in 27 of 40 tumours studied (Reveneau et al., 1999). Vakkala et al showed that carcinomas with both iNOS positive tumour and stromal cells had a higher apoptotic index and a higher calculated microvessel density index (Vakkala et al., 2000). Loibl et al further demonstrated that while none of the benign lesions were positive for iNOS, 67% *in situ* carcinomas and 61% invasive lesions showed iNOS tumour cell staining (Loibl et al., 2002).

In addition to breast cancer, iNOS has also been shown to be markedly expressed in approximately 60% of human adenomas and in 20-25% of colon carcinomas, while expression was either low or absent in the surrounding normal tissues (Ambs et al., 1998a). In human ovarian cancer, iNOS activity has been localized in tumour cells and not found in normal tissue (Thomsen et al., 1995). Other tumours that have demonstrated iNOS gene expression are brain, head and neck, esophagus, lung, prostate, bladder, pancreatic, and Kaposi's sarcoma (Cobbs et al., 1995; Rosbe et al., 1995; Wilson et al., 1998; Ambs et al., 1998a; Klotz et al., 1998; Hajri et al., 1998; Weninger et al., 1998; Swane et al., 1999).

In this study, we found that exist strong expression of iNOS in stroma and gland of HPB, in comparison with PCa. It Have been demonstrated that NO- mediated up-regulation of VEGF. In the results is possible that NO generated by iNOS in stroma may promote early new blood vessel formation by up-regulating VEGF and enhance ability of the tumour to grow and increases its invasiveness ability in gland. (Ambs et al., 1998a). Moreover, the accumulation of p53 in gland can result in down-regulation of iNOS expression by inhibition of iNOS promoter activity (Ambs et al., 1998b). On the other hand, the generation of chronic injury and irritation initiate the inflammatory response of stroma to gland (NOS-1). A subsequent respiratory burst an increase uptake of oxygen that leads to the release of reactive oxygen species (NO, ONOO , N2O3, NO2 and NO3) from leucocytes can damage surrounding cells and drive carcinogenesis by altering targets and pathways that are crucial to normal prostate homeostasis (Coussens & Werb, 2002; Fukumura et al., 2006).

COX-1 and COX-2 regulate a key step in prostanoid (i.e., tromboxanes and prostaglandins) synthesis. Prostaglandins regulate various pathophysiological processes such as inflammatory reaction, gastrointestinal cytoprotection and ulceration (Smith & Langenbach,

Tumoral Markers in Prostate Cancer 123

Our research group is determining the gene expression and activity of nitric oxide synthases isoforms (eNOS, nNOS and iNOS), Mn-SOD, Cu/Zn-SOD, Glutathione peroxidase, Glutathione reductase, Glutathione-S-transferase, Catalase and Ciclooxygenase-2 to integrate the effect of the regulation of the antioxidant system in the development of

Actually, we begin a new line of research where we studied the gene expression and polymorphisms of some components of the cytochrome P450 system as well as its association with the risk of developing prostate cancer and breast cancer. We found a protein over-expression of CYP2W1, 4F11 and 8A1, orphans cytochromes, in prostate cancer. We hope to find a molecular marker or prognostic indicator for prostate cancer and

[1] Alderton, W.K.; Cooper, C.E. & Knowles, R.G. (2001). Nitric oxide synthases: structure,

[2] Ambs, S.; Merriam, W.G.; Bennett, W.P.; Felley-Bosco, E.; Ogunfusika, M.O.; Oser, S.M.;

[3] Ambs, S.; Ogunfusika, M.O.; Merriam, W.G.; Bennett, W.P.; Billiar, T.R. & Harris, C.C.

[4] Arbiser, J.L.; Petros, J; Klafter, R.; Govindajaran, B.; McLaughlin, E.R.; Brown, L.F.;

[5] Arnold, R.S.; Shi, J.; Murad, E.; Whalen, A.M.; Sun, C.Q.; Polavarapu, R.; Parthasarathy,

[6] Babior, B.M. (1999). NADPH oxidase: an update. *Blood,* Vol. 92, No. 5, (March 1999), pp.

[7] Bánfi, B.; Molnár, G.; Maturana, A.; Steger, K.; Hegedûs, B.; Demaurex, N. & Krause,

[8] Bewick, M.A.; Conlon, M.S. & Lafrenie, R.M. (2008). Polymorphisms in manganese

function and inhibition. *Journal of Biochremistry,* Vol. 357, No. 3, (August 2001), pp.

Klein, S.; Shields, P.G.; Billiar, T.R. & Harris, C.C. (1998a) Frequent nitric oxide synthase-2 expression in human colon adenomas: implication for tumor angiogenesis and colon cancer progression. *Cancer Research*, Vol. 58, No. 2, (January

(1998b). Up-regulation of inducible nitric oxide synthase expression in cancerprone p53 knockout mice. *Proceedings of the National Academy of Sciences U S A,* Vol.

Cohen, C.; Moses, M.; Kilroy, S.; Arnold, R.S. & Lambeth, J.D. (2002). Reactive oxygen generated by Nox1 triggers the angiogenic switch. *Proceedings of the National Academy of Sciences U S A,* Vol. 99, No 2, (January 2002), pp. 715-720, ISSN 0027-

S.; Petros, .JA. & Lambeth, J.D. (2001). Hydrogen peroxide mediates the cell growth and transformation caused by the mitogenic oxidase Nox1. *Proceedings of the National Academy of Sciences U S A,* Vol. 98, No 10, (May 2001), pp. 5550-5555, ISSN

K.H. (2001). A Ca2+-activated NADPH oxidase in testis, spleen, and lymph nodes. The *Journal of Biological Chemistry,* Vol. 276, No. 40, (October 2001), pp. 37594-37601,

superoxide dismutase, myeloperoxidase and glutathione-*S-*transferase and survival

**6. Future research** 

breast cancer.

**7. References** 

8424

0027-8424

ISSN 0021-9258

1454–1476, ISSN 0006-4971

prostate cancer and recently in breast cancer.

593-615, ISSN 0021-924X

1998), pp. 334-341, ISSN 1538-7445

95, No. 15, (July 1998), pp. 8823-8828, ISSN 0027-8424

2001). COX-1 is the constitutive isoform and COX-2 is the inducible isoform. COX-1 is expressed in most tissues and plays a role in the production of prostaglandins that control normal physiological processes. COX-2 is undetectable in most normal tissues (except for the central nervous system, kidneys and seminal vesicles), but is induced by various inflammatory and mitogenic stimuli (growth factors, pro-inflammatory cytokines and tumor necrosis factor) and other regulatory factors (Peppelenbosch et al., 1993; Zhang et al., 1998, Chen et al., 2001; Dempke et al., 2001). Although the mechanism of COX-2 upregulation is not fully understood, it could result from activation of Ras and mitogen-activated protein kinase (MAPK) pathway. It has been recognized that Akt/PKB activity is implicated in Rasinduced expression of COX-2. COX-2 is regulated at transcriptional and post-transcriptional levels by proinflamatory agents. These pathways lead to the activation of regulatory factors that eventually bind the promoter region of the COX-2 gene. (Sheng et al., 1998, 2000).

In this study, we found that exist strong expression of COX-2 in stroma and gland of HPB, in comparison with PCa. There are conflicting data regarding whether COX-2 is increased in the epithelial , gland or the stromal component of tumors (Horsman et al., 2010). Liu et al were the first to describe tumorigenesis induced by COX-2 over-expression. In their study, the murine COX-2 gene was inserted downstream of a murine mammary tumor virus promoter. As a consequence, hyperplasia and carcinoma of the mammary gland were observed and associated with strong COX-2 expression in mammary gland epithelial cells with increase prostaglandin E2 levels. (Liu et al., 2001). The role of COX-2 in tumor promotion is more strongly supported by previous studies in colorectal tumor models describen by Oshima et al (Oshima et al., 1996). These findings have been confirmed analyzing many tumors including pancreas, skin, gastric, bladder, lung, head, and neck cancers, suggesting that COX-2, but not COX-1, may play a pivotal role in tumor formation and growth (Thun et al., 2002). COX-2-derived prostaglandins contribute to tumor growth by inducing angiogenesis that sustain tumor cell viability and growth. COX-2 is expressed within human tumor neovasculature as well as in neoplastic cells present in human colon, breast, prostate and lung cancer biopsy tissue. (Kerbel & Folkman, 2002). The proangiogenic effects of COX-2 are mediated primarily by three products of arachidonic metabolism: Tromboxane A2, Prostaglandins I2 and E2 and selective inhibition of COX-2 activity has been shown to suppress angiogenesis *in vitro* and *in vivo* (Tsujii et al., 1998; Masferrer et al., 2000; Uefuji et al., 2000). We suggested that COX-2 overexpression in stroma inhibit apoptosis and promote angiogenesis in prostate gland.

Our results suggest that iNOS and COX-2 play a key role in tumorigenesis and indicate that iNOS and COX-2-selective inhibitors could be a novel class of therapeutic agents for PCa.

#### **5. Conclusions**

We suggested that the O2 /H2O2 balance regulated by the over-expression of NOX, Cu/Zn-SOD, Mn-SOD and CAT is actively involved in tumor environment, cell proliferation, differentiation, tumor progression and angiogenesis of PCa. On the other hand, iNOS and COX-2 may promote blood vessel formation in gland from its overexpression in stroma by multiple mechanisms that involve reactive oxygen species, transcription factors, cytokines, growth factors and tumor necrosis factor.

Moreover, we suggested that the NOX, Cu/Zn-SOD, Mn-SOD, CAT, iNOS and/or COX-2 in combination with PSA, could be a molecular markers or prognostic indicators for the early diagnosis and post-treatment monitoring of PCa.

#### **6. Future research**

122 Prostate Cancer – Original Scientific Reports and Case Studies

2001). COX-1 is the constitutive isoform and COX-2 is the inducible isoform. COX-1 is expressed in most tissues and plays a role in the production of prostaglandins that control normal physiological processes. COX-2 is undetectable in most normal tissues (except for the central nervous system, kidneys and seminal vesicles), but is induced by various inflammatory and mitogenic stimuli (growth factors, pro-inflammatory cytokines and tumor necrosis factor) and other regulatory factors (Peppelenbosch et al., 1993; Zhang et al., 1998, Chen et al., 2001; Dempke et al., 2001). Although the mechanism of COX-2 upregulation is not fully understood, it could result from activation of Ras and mitogen-activated protein kinase (MAPK) pathway. It has been recognized that Akt/PKB activity is implicated in Rasinduced expression of COX-2. COX-2 is regulated at transcriptional and post-transcriptional levels by proinflamatory agents. These pathways lead to the activation of regulatory factors that eventually bind the promoter region of the COX-2 gene. (Sheng et al., 1998, 2000). In this study, we found that exist strong expression of COX-2 in stroma and gland of HPB, in comparison with PCa. There are conflicting data regarding whether COX-2 is increased in the epithelial , gland or the stromal component of tumors (Horsman et al., 2010). Liu et al were the first to describe tumorigenesis induced by COX-2 over-expression. In their study, the murine COX-2 gene was inserted downstream of a murine mammary tumor virus promoter. As a consequence, hyperplasia and carcinoma of the mammary gland were observed and associated with strong COX-2 expression in mammary gland epithelial cells with increase prostaglandin E2 levels. (Liu et al., 2001). The role of COX-2 in tumor promotion is more strongly supported by previous studies in colorectal tumor models describen by Oshima et al (Oshima et al., 1996). These findings have been confirmed analyzing many tumors including pancreas, skin, gastric, bladder, lung, head, and neck cancers, suggesting that COX-2, but not COX-1, may play a pivotal role in tumor formation and growth (Thun et al., 2002). COX-2-derived prostaglandins contribute to tumor growth by inducing angiogenesis that sustain tumor cell viability and growth. COX-2 is expressed within human tumor neovasculature as well as in neoplastic cells present in human colon, breast, prostate and lung cancer biopsy tissue. (Kerbel & Folkman, 2002). The proangiogenic effects of COX-2 are mediated primarily by three products of arachidonic metabolism: Tromboxane A2, Prostaglandins I2 and E2 and selective inhibition of COX-2 activity has been shown to suppress angiogenesis *in vitro* and *in vivo* (Tsujii et al., 1998; Masferrer et al., 2000; Uefuji et al., 2000). We suggested that COX-2 overexpression in stroma inhibit

Our results suggest that iNOS and COX-2 play a key role in tumorigenesis and indicate that iNOS and COX-2-selective inhibitors could be a novel class of therapeutic agents for PCa.

Cu/Zn-SOD, Mn-SOD and CAT is actively involved in tumor environment, cell proliferation, differentiation, tumor progression and angiogenesis of PCa. On the other hand, iNOS and COX-2 may promote blood vessel formation in gland from its overexpression in stroma by multiple mechanisms that involve reactive oxygen species,

Moreover, we suggested that the NOX, Cu/Zn-SOD, Mn-SOD, CAT, iNOS and/or COX-2 in combination with PSA, could be a molecular markers or prognostic indicators for the early

transcription factors, cytokines, growth factors and tumor necrosis factor.

/H2O2 balance regulated by the over-expression of NOX,

apoptosis and promote angiogenesis in prostate gland.

diagnosis and post-treatment monitoring of PCa.

**5. Conclusions** 

We suggested that the O2

Our research group is determining the gene expression and activity of nitric oxide synthases isoforms (eNOS, nNOS and iNOS), Mn-SOD, Cu/Zn-SOD, Glutathione peroxidase, Glutathione reductase, Glutathione-S-transferase, Catalase and Ciclooxygenase-2 to integrate the effect of the regulation of the antioxidant system in the development of prostate cancer and recently in breast cancer.

Actually, we begin a new line of research where we studied the gene expression and polymorphisms of some components of the cytochrome P450 system as well as its association with the risk of developing prostate cancer and breast cancer. We found a protein over-expression of CYP2W1, 4F11 and 8A1, orphans cytochromes, in prostate cancer. We hope to find a molecular marker or prognostic indicator for prostate cancer and breast cancer.

#### **7. References**


Tumoral Markers in Prostate Cancer 125

[21] Genkinger, J.M.; Platz, E.A.; Hoffman, S.C.; Strickland, P.; Huang, H.Y.; Comstock, G.W.

[22] Grigolo, B.; Lisignoli, G.; Toneguzzi, S.; Mazzetti, I. & Facchini, A. (1998). Copper/zinc

[23] Hajri, A.; Metzger, E.; Vallat, F.; Coffy, S.; Flatter, E.; Evrard, S.; Marescaux, J. &

[24] Hall, S.E.; Holman, C.D.; Wisniewsk, Z.S. & Semmens, J. (2004).Prostate cancer: socio-

[26] Horsman, M.R.; Bohn, A.B. & Busk, M. (2010). Vascular targeting therapy: potential

[27] Iwakiri, Y.; Satoh, A.; Chatterjee, S.; Toomre, D.K.; Chalouni, C.M.; Fulton, D.;

[28] Janssen, A.M.; Bosman, C.B.; van Duijn, W.; Oostendorp-van de Ruit, M.M.; Kubben,

[29] Kerbel, R. & Folkman, J. (2002). Clinical translation of angiogenesis inhibitors. *Nature Reviews Cancer,* Vol. 2, No. 10, (October 2002), pp. 727–739, ISSN 1474-175X [30] Klein, E.A.; Casey, G. & Silverman, R. (2006). Genetic susceptibility and oxidative tress

[31] Klotz, T.; Bloch, W.; Volberg, C.; Engelmann, U. & Addicks, K. (1998). Selective

[32] Lim, S.D.; Sun, C.; Lambeth, J.D.; Marshall, F.; Amin, M.; Chung, L.; Petros, J.A. &

*Prostate,* Vol. 62, No. 2, (February 2005), pp. 200–207, ISSN 0270-4137 [33] Liu, C.H.; Chang, S.H.; Narko, K.; Trifan, O.C.; Wu, M.T.; Smith, E.; Haudenschild, C.;

Vol. 127, No. 4, (April 2006), pp. 371-377, ISSN 0047-6374

18, No. 1, (February 2007), pp. 79-89, ISSN 0957-5243

No. 3, (September 2010), pp. 143-148, ISSN 1812-9269

No. 52, (December 2006), pp. 19777-19782, ISSN 0027-8424

68, No 6, (December 2006), .pp. 1145–1151, ISSN 0090-4295

Vol. 82, No. 10, (May 1998), pp. 1897-903, ISSN 1097-0142

276, No. 1, (January 2001), pp. 18563–18569, ISSN 0021-9258

2000), pp. 3183-3192, ISSN 1078-0432

0250-7005

849, ISSN 0007-0920.

& Helzlsouer, K.J. (2006). C47T polymorphism in manganese superoxide dismutase (MnSOD) antioxidant intake and survival. *Mechanisms of Ageing and Development,* 

superoxide dismutase expression by different human osteosarcoma cell lines. Anticancer Research, Vol. 18, No. 2A, (March-April 2008), pp. 1175-1180, ISSN

Aprahamian, M. (1998). Role of nitric oxide in pancreatic tumour growth: in vivo and in vitro studies. *British Journal of Cancer,* Vol. 78, No. 7, (October 1998), pp. 841-

economic, geographical and private-health insurance effects on care and survival *British Journal of Urology,* Vol.95, No. 1, (January 2005), pp. 51-58, ISSN 0007-1331 [25] Han, J.; Colditz, G.A. & Hunter, D.J. (2007). Manganese superoxide dismutase

polymorphism and risk of skin cancer (United States). *Cancer Causes Control,* Vol.

benefit depends on tumor and host related effects. *Experimental Oncology*, Vol. 32,

Groszmann, R.J.; Shah, V.H. & Sessa, W.C. (2006). Nitric oxide synthase generates nitric oxide locally to regulate compartmentalized protein S-nitrosylation and protein trafficking. *Proceedings of the National Academy of Sciences U S A*, Vol. 103,

F.J.; Griffioen, G.; Lamers, C.B.; van Krieken, J.H.; van de Velde, C.J. & Verspaget, H.W. (2000). Superoxide dismutases in gastric and esophageal cancer and the prognostic impact in gastric cancer. *Clinical Cancer Research*, Vol. 6, No. 8, (August

in prostate cancer: integrated model with implications for prevention. *Urology,* Vol.

expression of inducible nitric oxide synthase in human prostate carcinoma. *Cancer,*

Arnold, R.S. (2005) Increased Nox1 and hydrogen peroxide in prostate cancer.

Lane, T.F. & Hla, T. (2001). Over-expression of cyclooxygenase-2 is sufficient to induce tumorigenesis in transgenic mice. The *Journal of Biological Chemistry*, Vol.

after treatment for metastatic breast cancer. *Breast Cancer Research and Treatment,* Vol. 111, No. 1, (September 2008), pp. 93-101, ISSN 0167-6806


[9] Brar, S.S.; Corbin, Z.; Kennedy, T.P.; Hemendinger, R.; Thornton, L.; Bommarius, B.;

[10] Brown, N.M.; Torres, A.S.; Doan Pe, & O`Halloran, T.V. (2004). Oxygen and the copper

[11] Chen, C.C.; Sun, Y.T.; Chen, J.J. & Chang, Y.J. (2001). Tumor necrosis factor-induced

[12] Church, J.E. & Fulton, D. (2006). Differences in eNOS activity because of subcellular

[13] Cobbs, C.S.; Brenman, J.E.; Aldape, K.D.; Bredt, D.S. & Israel, MA. (1995). Expression of

[14] Coussens, L.M. & Werb, Z. (2002). Inflammation and cancer. *Nature,* Vol. 420, No. 6917,

[15] Dannenberg, A.J. & Zakim, D. (1999). Chemoprevention of colorectal cancer through

[16] Dempke, W.; Rie, C.; Grothey, A. & Schmoll, H.J. (2001). Cyclooxygenase-2: a novel

[17] Eberhart, C.E.; Coffey, R.J.; Radhika, A.; Giardiello, F.M.; Ferrenbach, S. & DuBois, R.N.

[18] Forsberg, L.; Lyrenas, L.; de Faire, U. & Morgenstern, R. (2001). A common functional

[19] Friebe, A. & Koesling, D. (2003). Regulation of nitric oxide-sensitive guanylyl cyclase. *Circulation Research,* Vol. 93, No. 2, (July 2003), pp. 96-105, ISSN 00097330 [20] Fukumura, D.; Kashiwagi, S. & Jain, R.K. (2006). The role of nitric oxide in tumour

55, No. 4, (February 1995), pp. 727-730, ISSN 1538-7445

Vol. 127, No. 7, (July 2001), pp. 411–417, ISSN 0171-5216

(December 2002), pp. 860-867, ISSN 0028-0836

Vol. 111, No. 1, (September 2008), pp. 93-101, ISSN 0167-6806

(April 2004), pp. 5518–5523, ISSN 0027-8424

ISSN 0363-6143

1477-1488, ISSN 0021-9258

pp. 499–504, ISSN 0093-7754

pp. 1183–1188, ISSN 0016-5085

pp. 500 –505, ISSN 0891-5849

1474-175X

895X

after treatment for metastatic breast cancer. *Breast Cancer Research and Treatment,*

Arnold, R.S.; Whorton, A.R.; Sturrock, A.B.; Huecksteadt, T.P.; Quinn, M.T.; Krenitsky, K.; Ardie, K.G.; Lambeth, J.D. & Hoidal, J.R. (2003). NOX5 NAD(P)H oxidase regulates growth and apoptosis in DU 145 prostate cancer cells. *American Journal of Physiology.Cell Physiology,* Vol. 285, No. 2, (August 2003), pp. C353-C369,

chaperone CCS regulate posttranslational activation of Cu, Zn superoxide dismutase. *Proceedings of the National Academy of Sciences U S A,* Vol. 101, No. 15,

cyclooxygenase-2 expression via sequential activation of ceramide-dependent mitogenactivated protein kinases, and I*\_*B kinase 1/2 in human alveolar epithelial cells. *Molecular Pharmacology* Vol. 59, No. 3, (March 2001), pp. 493–500, ISSN 0026-

localization are dictated by phosphorylation state rather than the local calcium environment. *The Journal of Biological Chemistry,* Vol. 281, No. 3, (October 2005), pp.

nitric oxide synthase in human central nervous system tumors. *Cancer Research*, Vol.

inhibition of cyclooxygenase-2. *Seminars in Oncology*, Vol. 26, No. 5, (October 1999),

target for cancer chemotherapy? *The Journal of Cancer Research of Clinical On*cology,

(1994) Up-regulation of cyclooxygenase 2 gene expression in human colorectal adenomas and adenocarcinomas. *Gastroenterology*, Vol. 107, No. 4, (October 1994),

C-T substitution polymorphism in the promoter region of the human catalase gene influences transcription factor binding, reporter gene transcription and is correlated to blood catalase levels. *Free Radical Biology & Medicine,* Vol. 30, No. 5, (March 2001),

progression. *Nature Reviews Cancer*, Vol. 6, No. 7, (July, 2006), pp. 521-534, ISSN


Tumoral Markers in Prostate Cancer 127

[46] Smith, W.L. & Langenbach, R. (2001). Why there are two cyclooxygenase isozymes. *The* 

[47] Sonoshita, M.; Takaku, K.; Sasaki, N.; Sugimoto, Y.; Ushikubi, F.; Narumiya, S. &

[48] Stamler, J.S. (1994). Redox signaling: nitrosylation and related target interactions of nitric oxide. *Cell,* Vol. 78, No. 3, (September 1994), pp. 931-936, ISSN 0092-8674 [49] Starcevic, S.L.; Diotte, N.M.; Zukowski, K.L.; Cameron, M.J. & Novak, R.F. (2003).

[50] Stuehr, D.J.; Santolini, J.; Wang, Z.Q.; Wei, C.C. & Adak, S. (2004). Update on

[53] Thun, M.J.; Henley, S.J. & Patrono, C. (2002). Nonsteroidal anti-inflammatory drugs as

[54] Tsujii, M.; Kawano, S.; Tsuji, S.; Sawaoka, H; Hori, M. & DuBois, R.N. (1998).

[55] Uefuji, K.; Ichikura, T. & Mochizuki, H. (2000). Cyclooxygenase-2 expression is related

*Cancer Research*, Vol. 6, No. 1, (January 2000), pp. 135–138, ISSN 1557-3265 [56] Tugcu, V.; Ozbek, E.; Aras, B.; Arisan, S.; Caskurlu, T. & Tasci, A.I. (2007). Manganese

[57] Uefuji, K.; Ichikura, T. & Mochizuki, H. Cyclooxygenase-2 expression is related to

*Cancer Research*, Vol. 6, No. 1, (January 2000), pp. 135–138, ISSN 1557-3265 [58] Ushio-Fukai, M. & Nakamura, Y. (2008). Reactive oxygen species and angiogenesis:

[59] Vakkala, M.; Kahlos, K.; Lakari, E.; Paakko, P.; Kinnula, V. & Soini, Y. (2000). Inducible

*Research,* Vol. 35, No 5, (October 2007), pp. 219-224, ISSN 0300-5623

*Chemistry*, Vol. 279, No. 35, (May 2004), pp. 27257-27562, ISSN 0021-9258 [51] Swana, H.S.; Smith, S.D.; Perrotta, P.L.; Saito, N.; Wheeler, M.A. & Weiss, R.M. (1999).

*Cancer*, Vol. 72, No. 1, (July 1995), pp. 41-44, ISSN 0007-0920

No. 5, (May 1998), pp. 705–716, ISSN 0092-8674

2008), pp. 37-52, ISSN 0304-3835

2408-2416, ISSN 1557-3265

6635, ISSN 0021-9258

2001), pp. 1048–1051, ISSN 1078-8956

pp. 74-81, ISSN 1096-6080

0021-9738

mRNA. The *Journal of Biological Chemistry*, Vol. 275, No. 9, (March 2000), pp. 6628–

*Journal of Clinical Investigation*, Vol. 107, No. 12, (June 2001), pp. 1491–1495, ISSN

Oshima, M. (2001). Acceleration of intestinal polyposis through prostaglandin receptor EP2 in Apc-knockout mice. *Nature Medicine,* Vol. 7, No. 9, (September

Oxidative DNA damage and repair in a cell lineage model of human proliferative breast disease (PBD). *Toxicological Sciences (U.S.),* Vol. 75, No.1, (September 2003),

mechanism and catalytic regulation in the NO synthases. *The Journal of Biological* 

Inducible nitric oxide synthase with transitional cell carcinoma of the bladder. *Journal of Urology*, Vol. 161, No. 2, (February 1999), pp. 630-634, ISSN 0022-5347 [52] Thomsen, L.L.; Miles, D.W.; Happerfield, L.; Bobrow, L.G.; Knowles, R.G. & Moncada,

S. (1995). Nitric oxide synthase activity in human breast cancer. *British Journal of* 

anticancer agents: mechanistic, pharmacologic, and clinical issues," *Journal of the National Cancer Institute*, Vol. 94, No. 4, (February 2002), pp. 252–266, ISSN 0027-8874

Cyclooxygenase regulates angiogenesis induced by colon cancer cells. *Cell,* Vol. 93,

to prostaglandin biosynthesis and angiogenesis in human gastric cancer. *Clinical* 

superoxide dismutase (Mn-SOD) gene polymorphisms in urolithiasis. *Urological* 

prostaglandin biosynthesis and angiogenesis in human gastric cancer. *Clinical* 

NADPH oxidase as target for cancer therapy. *Cancer Letters*, Vol. 266, No. 1, (July

nitric oxide synthase expression, apoptosis, and angiogenesis in in situ and invasive breast carcinomas. *Clinical Cancer Research*, Vol. 6, No. 6, (June 2000), pp.


[34] Liu, G.; Zhou, W.; Park, S.; Wang, L.I.; Miller, D.P.; Wain, J.C.; Lynch, T.J.; Su, L. &

[35] Loibl, S.; von Minckwitz, G.; Weber, S.; Sinn, H.P.; Schini-Kerth, V.B; Lobysheva, I.;

[36] López Laur, J.D.; Abud, M.; López Fontana, C.; Silva, J.; Cisella, Y.; Pérez Elizalde, R. &

*Españoles de Urología*, Vol. 61, No. 5, (June 2008), pp. 563-569, ISSN 0004-0614 [37] Masferrer, J.L.; Leahy, K.M.; Koki, A.T.; Zweifel, B.S.; Settle, S.L.; Woerner, B.M.;

[38] Oess, S.; Icking, A.; Fulton, D.; Govers, R. & Müller-Esterl, W. (2006). Subcellular

[39] Oshima, M.; Dinchuk , J.E.; Kargman, S.L.; Oshima, H.; Hancock, B.; Kwong, E.;

[40] Peppelenbosch, M.P.; Tertoolen, L.G.; Hage, W.J. & de Laat, S.W. (1993). Epidermal

[41] Qian, Y.; Zheng, Y.; Abraham, L.; Ramos, K.S. & Tiffany-Castiglioni, E. (2005).

[42] Reveneau, S.; Arnould, L.; Jolimoy, G.; Hilpert, S.; Lejeune, P.; Saint-Giorgio, V.;

[43] Rosbe, K.W.; Prazma, J.; Petrusz, P.; Mims, W.; Ball, S.S. & Weissler, M.C. (1995).

[45] Sheng, H.; Shao, J.; Dixon, D.A.; Williams, C.S.; Prescott, S.M.; DuBois, R.N. &

*Surgery*, Vol.113, No. 5, (November 1995), pp. 541-549, ISSN 01945998 [44] Sheng, H.; Shao, J. & Dubois, R.N. (2001). K-Ras-mediated increase in cyclooxygenase 2

No. 6, (March 2001), pp. 2670–2675, ISSN 1538-7445

Vol. 87, No. 5, (November 1996), pp. 803–809, ISSN 0092-8674

(December 2004), pp. 2802-2808, ISSN 0008-543X

No. 5, (March 2000), pp. 1306–1311, ISSN 1538-7445

No. 3, (June 2006) pp. 401-409, ISSN 0021-924X

2005), pp. 323-332, ISSN 0169-328X

0092-8674

ISSN 0023-6837

Christiani, D.C. (2004). The SOD2 Val/Val genotype enhances the risk of non small cell lung carcinoma by p53 and XRCC1 polymorphisms. *Cancer,* Vol. 101, No. 12,

Nepveu, F.; Wolf, G.; Strebhardt, K. & Kaufmann, M. (2002). Expression of endothelial and inducible nitric oxide synthase in benign and malignant lesions of the breast and measurement of nitric oxide using electron paramagnetic resonance spectroscopy. *Cancer* , Vol. 95, No. 6, (September 2002), pp. 1191-1198, ISSN 1097-0142

Ortiz, A. (2008). Antioxidant power and cellular damage in prostate cancer. *Archivos* 

Edwards, D.A.; Flickinger, A.G.; Moore, R.J. & Seibert, K. (2000). Antiangiogenic and antitumor activities of cyclooxygenase-2 inhibitors. *Cancer Research*, Vol. 60,

targeting and trafficking of nitric oxide synthases. *Journal of Biochremistry*, Vol. 396,

Trzaskos, J.M.; Evans, J.F. & Taketo, M.M. (1996). Suppression of intestinal polyposis in *Apc*-knockout mice by inhibition of cyclooxygenase-2 (COX-2). *Cell,*

growth factor-induced actin remodeling is regulated by 5-lipoxygenase and cyclooxygenase products. *Cell*, Vol. 74, No. 3, (August, 1993), pp. 565–575, ISSN

Differential profiles of copper-induced ROS generation in human neuroblastoma and astrocytoma cells. *Brain Research. Molecular Brain Research*, Vol. 134, No 2, (April

Belichard, C. & Jeannin, J.F. (1999). Nitric oxide synthase in human breast cancer is associated with tumor grade, proliferation rate, and expression of progesterone receptors. Laboratory Investigation, Vol. 79, No. 10, (October 1999), pp. 1215-25,

Immunohistochemical characterization of nitric oxide synthase activity in squamous cell carcinoma of the head and neck. *Otolaryngology: Head and Neck* 

mRNA stability involves activation of the protein kinase B. *Cancer Research*, Vol. 61,

Beauchamp, R.D. (2000). Transforming growth factor-1 enhances Ha-ras-induced expression of cyclooxygenase-2 in intestinal epithelial cells via stabilization of mRNA. The *Journal of Biological Chemistry*, Vol. 275, No. 9, (March 2000), pp. 6628– 6635, ISSN 0021-9258


**9** 

*1Algérie 2Saudi Arabia* 

> *3,4UK 5Qatar*

**Improving Prostate Cancer Classification:** 

*1Département d'Electronique, Faculté des Sciences de l'Ingénieur, Université de Jijel, Jijel* 

Over the last decade prostate cancer has become one of the most common cancer in male population with an estimated 1.37 million people diagnosed and 200,000 annual death rate worldwide ( Stewart & Kleihues, 2003). Biopsies are often advised after a Prostate Specific Antigen (PSA) test reveals high levels of PSA in the blood which usually indicate high risks of Prostatic Carcinoma (PCa). The biopsy is needed because high PSA levels can also be caused by other benign conditions like Benign Prostatic Hyperplasia (BPH) (Kronz, Westra

Biopsy of the prostate, usually stained by Hematoxylin and Eosin (H&E) technique, is the key step for confirming the diagnosis of malignancy and grading treatment. By viewing the microscopic images of biopsy specimens, pathologists can determine the histological grades. In December 1999, a study of more than 6,000 patients by Johns Hopkins researchers found that up to two out of every 100 people who come to larger medical centers for treatment, following a biopsy, are given a diagnosis that is "totally wrong". The results suggested that second opinion pathology examinations not only prevent errors, but also save lives and money. Human assessment is time consuming and very subjective due to inter- and intraobserver variations. At present, most diagnosis of cancer is still done by visual examination of radiological images, microscopy of biopsy specimens, direct observation and so on. These views are typically interpreted in a qualitative manner by clinicians trained to classify abnormal features such as structural irregularities. A more quantitative and reproducible approach for analyzing images is highly desired. Therefore, how to develop a more

**1. Introduction** 

& Epstein, 1999)

**A Round Robin Forward Sequential** 

**Selection Approach** 

Sabrina Bouatmane1, Ahmed Bouridane2,3,

Mohamed Ali Roula4 and Somaya Al-Maadeed5

*2Departemt of Computer Science, King Saud University, Riyadh 3School of Computing, Engineering and Information Sciences Northumbria University at Newcastle, Pandon Building* 

*4Faculty of Advanced Technology, University of Glamorgan, Pontypridd 5Department of Computer Science & Engineering Qatar University, Doha* 


### **Improving Prostate Cancer Classification: A Round Robin Forward Sequential Selection Approach**

 Sabrina Bouatmane1, Ahmed Bouridane2,3, Mohamed Ali Roula4 and Somaya Al-Maadeed5 *1Département d'Electronique, Faculté des Sciences de l'Ingénieur, Université de Jijel, Jijel 2Departemt of Computer Science, King Saud University, Riyadh 3School of Computing, Engineering and Information Sciences Northumbria University at Newcastle, Pandon Building 4Faculty of Advanced Technology, University of Glamorgan, Pontypridd 5Department of Computer Science & Engineering Qatar University, Doha 1Algérie 2Saudi Arabia 3,4UK 5Qatar* 

#### **1. Introduction**

128 Prostate Cancer – Original Scientific Reports and Case Studies

[60] Vignais, P.V. (2002). The superoxide-generating NADPH oxidase: structural aspects and

[61] Weninger, W.; Rendl, M.; Pammer, J.; Mildner, M.; Tschugguel, W.; Schneeberger, C.;

[63] Xia, C.; Meng, Q.; Liu, L.Z.; Rojanasakul, Y.; Wang, X.R. & Jiang, B.H. (2007). Reactive

[64] Xu, W. & Liu, L. (1998). Nitric Oxide: from a mysterious labile factor to the molecule of

[65] Zhang, F.; Subbaramaiah, K.; Altorki, N. & Dannenberg, A.J. (1998). Dihydroxy bile

*Chemistry*, Vol. 273, No. 4, (January 1998), pp. 2424-2428, ISSN 0021-9258 [66] Zhao, Y.; Kiningham, K.K.; Lin, S.M. & St Clair, D.K. (2001). Overexpression of MnSOD

*Investigation*, Vol. 78, No. 8, (August 1998), pp. 949-955, ISSN 0023-6837 [62] Wilson, K.T.; Fu, S.; Ramanujam, K.S. & Meltzer, S.J. (1998). Increased expression of

2002), pp. 1428–1459, ISSN 1420-682X

2934, ISSN 1538-7445

10823-10830, ISSN 0008-5472

pp. 375–386, ISSN 1523-0864

(December 1998), pp. 251-258, ISSN 1001-0602

activation mechanism. *Cell and Molecular Life Sciences,* Vol. 59, No. 9, (September

Stürzl, M. & Tschachler, E. (1998). Nitric oxide synthases in Kaposi sarcoma are expressed predominantly by vessels and tissue macrophages. *Laboratory* 

inducible nitric oxide synthase and cyclooxygenase-2 in Barrett s esophagus and associated adenocarcinomas. *Cancer Research*, Vol. 58, No. 14, (July 1998), pp.:2929-

oxygen species regulate angiogenesis and tumor growth through vascular endothelial growth factor. *Cancer Research*, Vol. 67, No. 22, (November 2007), pp.

the Nobel Prize. Recent progress in nitric oxide research. *Cell Research*, Vol. 8, No. 4,

acids activate the transcription of cyclooxygenase-2. The *Journal of Biological* 

protects murine fibrosarcoma cells (FSa-II) from apoptosis and promotes a differentiation program upon treatment with 5-azacytidine: involvement of MAPK and NFkappaB pathways. *Antioxidants & Redox Signaling,* Vol. 3, No. 3, (June 2001),

> Over the last decade prostate cancer has become one of the most common cancer in male population with an estimated 1.37 million people diagnosed and 200,000 annual death rate worldwide ( Stewart & Kleihues, 2003). Biopsies are often advised after a Prostate Specific Antigen (PSA) test reveals high levels of PSA in the blood which usually indicate high risks of Prostatic Carcinoma (PCa). The biopsy is needed because high PSA levels can also be caused by other benign conditions like Benign Prostatic Hyperplasia (BPH) (Kronz, Westra & Epstein, 1999)

> Biopsy of the prostate, usually stained by Hematoxylin and Eosin (H&E) technique, is the key step for confirming the diagnosis of malignancy and grading treatment. By viewing the microscopic images of biopsy specimens, pathologists can determine the histological grades. In December 1999, a study of more than 6,000 patients by Johns Hopkins researchers found that up to two out of every 100 people who come to larger medical centers for treatment, following a biopsy, are given a diagnosis that is "totally wrong". The results suggested that second opinion pathology examinations not only prevent errors, but also save lives and money. Human assessment is time consuming and very subjective due to inter- and intraobserver variations. At present, most diagnosis of cancer is still done by visual examination of radiological images, microscopy of biopsy specimens, direct observation and so on. These views are typically interpreted in a qualitative manner by clinicians trained to classify abnormal features such as structural irregularities. A more quantitative and reproducible approach for analyzing images is highly desired. Therefore, how to develop a more

Improving Prostate Cancer Classification:

PIN. (d) PCa.

A Round Robin Forward Sequential Selection Approach 131

and 2D spatial analysis. They have compared the results using a textural analysis on single hyperspectral band against 3D spectral spatial analysis of histological colon images. However, the classification features were not extracted from multispectral data but rather from segmented 2D images obtained from multispectral data. Roula et al have described a novel approach, in which additional spectral data is used for the classification of prostate needle

Fig. 1. Images showing representative samples of the four classes. (a) Stroma. (b) BPH. (c)

then we need at least 100 training samples per class to design a satisfactory classifier.

Another way to reduce the dimensionality of the feature space is by using feature selection methods. The term feature selection refers to the selection of the best subset of the input

The major problem arising in using multispectral data is high-dimensional feature vector size (> 100). The number of training samples used to design the classifier is small relative to the number of features. For such a high dimensionality problem, pattern recognition techniques suffer from the well-known curse-of-dimensionality (Jain et al., 2000): keeping the number of training samples limited and increasing the number of features will eventually result in badly performing classifiers. One way to overcome this problem is to reduce the dimensionality of the feature space. While a precise relationship between the number of training samples and the number of features is hard to establish, a combination of theoretical and empirical studies has suggested the following rule of thumb regarding the ratio of the sample size to dimensionality: the number of training samples per class should be greater than or equal to five times the features used (Dash & Liu, 1997). For example, if we have a feature vector of dimension 20,

biopsies (Roula, 2002, 2003) which reduced overall error rate from 11.6% to 5.1%.

objective computer-aided technique to automatically and correctly classify prostatic carcinoma is the goal of this research study. The aim here is to use automatic classifiers as a diagnosis aid along with human expertise by applying image processing and computer vision techniques to perform quantitative measurements of relevant features that can discriminate between different types of tissues that occur in biopsies. In the case of the prostate gland, four major classes of tissues have to be recognized and labeled by the pathologist (Figure 1 shows some samples of each class):


Numerous investigations have been carried out using different approaches such as morphology, texture analysis, and others for the classification of prostatic samples (Bartels et al., 1998; Clark et al., 1987). The Gleason grading system (Gleason & Tannenbaum, 1977) is a well known method. In this grading system, the prostate cancer can be classified into five tumor grades represented by a number ranging from 1 to 5 with five being the worst grade possible ( O'Dowd et al., 2001). Tabech et al. proposed (Tabesh et al., 2005) an automatic two-stage system for prostate cancer diagnosis with the Gleason grading. The color, morphometric and texture features are extracted from prostate tissue images in their system. Then, linear and quadratic Gaussian classifiers were used to classify images into tumor/non tumor classes and further categorized into low/high grades for cancer images. Huang et al. proposed (Huang & Lee, 2009) two feature methods based on fractal dimension to analyze the variations of intensity and texture complexity in the regions of interest. Each image can be classified into an appropriate grade by using Bayesian, KNN, and support vector machine (SVM) classifiers, respectively. Leave on out and k fold cross-validation procedures were used to estimate the correct classification rates.

However, all these studies have been performed using a color space that is limited either to gray-level images, or to the standard RGB channels. In both cases, the color sampling process results in a loss of a considerable amount of spectral information, which may be extremely valuable in the classification process. High throughput liquid crystal tunable filters (LCTF) have recently been used in pathology, enabling a complete high resolution optical spectrum to be generated at every pixel of a microscope image. Studies suggest that multispectral images can capture relevant data not present in conventional RGB images. In (Liu et al., 2002) the authors used a large set of multispectral texture features for the detection of cervical cancer. In (Barshack et al., 1999), spectral morphometric characteristics were used on specimen of breast carcinoma cells stained with haematoxylin and eosin (H&E). Their analysis showed a correlation between specific patterns of spectra and different groups of breast carcinoma cells. Larsh et al. (Larsh et al., 2002) suggested that multispectral imaging can improve the analysis of pathological scenes by capturing patterns that are transparent both to the human eye and the standard RGB imaging.

In ( Boucheron et al., 2007) Boucheron et al a comparison is performed between multispectral and RGB data for nuclei classification of breast tissue. Using SVM classifiers, the authors have concluded that multispectral bands do not contain much more discriminatory spectral information than the RGB bands for nuclei classification. However, the research was concerned with the classification of single pixels and it was limited to the classification of nuclei of histological breast images. Masood & Rajpoot (Masood & Rajpoot, 2008) present a study based on the comparison of two approaches: 3D spectral/spatial analysis

objective computer-aided technique to automatically and correctly classify prostatic carcinoma is the goal of this research study. The aim here is to use automatic classifiers as a diagnosis aid along with human expertise by applying image processing and computer vision techniques to perform quantitative measurements of relevant features that can discriminate between different types of tissues that occur in biopsies. In the case of the prostate gland, four major classes of tissues have to be recognized and labeled by the

pathologist (Figure 1 shows some samples of each class):

2. *benign prostatic hyperplasia*: BPH (a benign condition);

were used to estimate the correct classification rates.

3. *prostatic intraepithelial neoplasia*: PIN (a precursor state for cancer);

4. *prostatic carcinoma*: PCa (abnormal tissue development corresponding to cancer).

Numerous investigations have been carried out using different approaches such as morphology, texture analysis, and others for the classification of prostatic samples (Bartels et al., 1998; Clark et al., 1987). The Gleason grading system (Gleason & Tannenbaum, 1977) is a well known method. In this grading system, the prostate cancer can be classified into five tumor grades represented by a number ranging from 1 to 5 with five being the worst grade possible ( O'Dowd et al., 2001). Tabech et al. proposed (Tabesh et al., 2005) an automatic two-stage system for prostate cancer diagnosis with the Gleason grading. The color, morphometric and texture features are extracted from prostate tissue images in their system. Then, linear and quadratic Gaussian classifiers were used to classify images into tumor/non tumor classes and further categorized into low/high grades for cancer images. Huang et al. proposed (Huang & Lee, 2009) two feature methods based on fractal dimension to analyze the variations of intensity and texture complexity in the regions of interest. Each image can be classified into an appropriate grade by using Bayesian, KNN, and support vector machine (SVM) classifiers, respectively. Leave on out and k fold cross-validation procedures

However, all these studies have been performed using a color space that is limited either to gray-level images, or to the standard RGB channels. In both cases, the color sampling process results in a loss of a considerable amount of spectral information, which may be extremely valuable in the classification process. High throughput liquid crystal tunable filters (LCTF) have recently been used in pathology, enabling a complete high resolution optical spectrum to be generated at every pixel of a microscope image. Studies suggest that multispectral images can capture relevant data not present in conventional RGB images. In (Liu et al., 2002) the authors used a large set of multispectral texture features for the detection of cervical cancer. In (Barshack et al., 1999), spectral morphometric characteristics were used on specimen of breast carcinoma cells stained with haematoxylin and eosin (H&E). Their analysis showed a correlation between specific patterns of spectra and different groups of breast carcinoma cells. Larsh et al. (Larsh et al., 2002) suggested that multispectral imaging can improve the analysis of pathological scenes by capturing patterns

In ( Boucheron et al., 2007) Boucheron et al a comparison is performed between multispectral and RGB data for nuclei classification of breast tissue. Using SVM classifiers, the authors have concluded that multispectral bands do not contain much more discriminatory spectral information than the RGB bands for nuclei classification. However, the research was concerned with the classification of single pixels and it was limited to the classification of nuclei of histological breast images. Masood & Rajpoot (Masood & Rajpoot, 2008) present a study based on the comparison of two approaches: 3D spectral/spatial analysis

that are transparent both to the human eye and the standard RGB imaging.

1. *stroma*: STR (normal muscular tissue);

and 2D spatial analysis. They have compared the results using a textural analysis on single hyperspectral band against 3D spectral spatial analysis of histological colon images. However, the classification features were not extracted from multispectral data but rather from segmented 2D images obtained from multispectral data. Roula et al have described a novel approach, in which additional spectral data is used for the classification of prostate needle biopsies (Roula, 2002, 2003) which reduced overall error rate from 11.6% to 5.1%.

Fig. 1. Images showing representative samples of the four classes. (a) Stroma. (b) BPH. (c) PIN. (d) PCa.

The major problem arising in using multispectral data is high-dimensional feature vector size (> 100). The number of training samples used to design the classifier is small relative to the number of features. For such a high dimensionality problem, pattern recognition techniques suffer from the well-known curse-of-dimensionality (Jain et al., 2000): keeping the number of training samples limited and increasing the number of features will eventually result in badly performing classifiers. One way to overcome this problem is to reduce the dimensionality of the feature space. While a precise relationship between the number of training samples and the number of features is hard to establish, a combination of theoretical and empirical studies has suggested the following rule of thumb regarding the ratio of the sample size to dimensionality: the number of training samples per class should be greater than or equal to five times the features used (Dash & Liu, 1997). For example, if we have a feature vector of dimension 20, then we need at least 100 training samples per class to design a satisfactory classifier.

Another way to reduce the dimensionality of the feature space is by using feature selection methods. The term feature selection refers to the selection of the best subset of the input

Improving Prostate Cancer Classification:

**2.1 Texture feature** 

Angular Second Moment

Dissimilarity

Correlation

Where ,,, *x*

Entropy or randomness

Inverse difference Moment:

the co-occurrence matrix respectively.

Contrast or difference moment

A Round Robin Forward Sequential Selection Approach 133

To identify prostatic patterns, texture features are needed as a discriminative measurement for the samples. Haralick (Haralick, 1979) assumed that texture information is sufficiently identified by a matrix indexed by grey levels and where the elements represent the frequency of having two defined grey levels separated by a defined distance in a defined

> *co i j d* (, , , )

The above equation means that there are α pairs of pixels having i and j respectively, as grey

GLCM is computed, depend on the nature of the texture. Small d values are suitable for fine

For an image of 256 grey levels (Ng=256), there would be 65536 feature elements to use as a measure for the texture. Therefore, the direct use of the co-occurrence matrix is computationally intensive and as such is not practical. Instead, the texture features are represented by deriving some more meaningful measurements. A set of features was proposed by Haralick to characterise the homogeneity, the coarseness, the periodicity and

1 1

*i j ASM P i j* 

1 1

1 1

*i j DIS i j P i j* 

1 1

1 1

*i j*

*i j Pi j IDM*

*N Ng g*

*i j ENT P i j P i j* 

*N Ng g*

*COR i j P i j* 

1 1

*i j*

*N Ng g*

*N Ng g*

*i j CON i j P i j* 

*N Ng g*

*N Ng g*

2

(2)

(3)

(4)

(, )

2

( ) (, )

(, )

( )( ) ( , ) /( )

*<sup>y</sup> <sup>x</sup> <sup>y</sup>* are the means and the variances of the row sums and column sums of

( , )log (, )

(, ) (1 ( ) )

2

*x y x y*

2

 

(6)

(7)

(5)

(1)

]. The values of d, for which the

direction. This matrix is called grey level co-occurrence matrix (GLCM):

textures, whereas larger distances are needed to measure coarse textures.

levels and separated by the cylindrical co-ordinate [*d*,

the linearity of textures. These features are defined as follows:

feature set. These methods used in the design of pattern classifiers have three goals: (1) to reduce the cost of extracting the features, (2) to improve the classification accuracy, and (3) to improve the reliability of the estimation of the performance, since a reduced feature set requires less training samples in the training procedure of a pattern classifier (Jain et al., 2000). Feature selection produces savings in the measuring features (since some of the features are discarded) and the selected features retain their original physical interpretation.

In previous papers (Bouatmane et al., 2007), we addressed the high input dimensionality problem by selecting the best-subset of sequential forward selection SFS followed by a classification using a nearest neighbour classifier (1NN) technique. Although, this approach produced results superior to previously reported methods (Roula, 2002,2003) , the classification accuracy can be further improved by decomposing this multiclass problem into a number of simpler two-class problems. In this case, each subproblem can be regarded separately and solved using a suitable binary classifier. The outputs of this collection of classifiers can then be combined to produce the overall result for the original multiclass problem. In this paper, we propose a Round-Robin (RR) classification algorithm using a sequential forward selection/nearest neighbor (SFS/1NN) classifier to improve the classification accuracy. Round Robin classification is a technique which is suitable for use in multiclass problems. The technique consists of dividing the multiclass problem into an appropriate number of simpler binary classification problems (Furnkranz, 2002). Each binary classifier is implemented as an SFS/1NN classifier, and the final outcome is computed using majority voting technique. A key characteristic of this approach is that, in a binary class, the classifier attempts to find the features that only distinguish that particular class. Thus, different features are selected for each binary classifier, resulting in an overall increase in the classification accuracy. In contrast, in a multiclass problem, the classifier tries to find those features that distinguish all classes at once.

The remainder of this chapter is organized as follows: Sect. 2 gives a description of the dataset used including texture and structural features. Section 3 is concerned with feature selection problem and describes the RR approach followed by the probability estimate for the classifier outputs and error estimation. Sect. 4 describes the image acquisition and dataset. Sect. 5 gives the results obtained and their analysis and discussion including a performance comparative study. Sect. 6 analyses the features selected and sect. 7 gives ROC curves and finally sect. 8 gives a summary of the chapter.

#### **2. Images features**

Over the last years, the most prolific and promising works in the area of cancer classification have been in the area of texture analysis of the nucleus (Tabesh et al., 2005; Liu et al., 20). This is not surprising since pre-cancerous abnormalities are manifested in visual and subvisual changes in cell characteristics. In fact, it is generally believed that the initial signs of cell neoplasia appear in the nucleus. Because nuclear chromatin and its spatial arrangement can be viewed as a type of texture and whether tissue samples are examined at low, medium or high magnification, texture is a key element in the differentiation between normal and malignant tissue patterns.

However, texture features are not sufficient to classify all the groups. The complex structures present in BPH, PIN and also PCa need a higher level description. Thus, structural features, based on segmentation, have been computed for different spectral bands and consolidated in a large feature vector. The features used are described in the following subsections.

#### **2.1 Texture feature**

132 Prostate Cancer – Original Scientific Reports and Case Studies

feature set. These methods used in the design of pattern classifiers have three goals: (1) to reduce the cost of extracting the features, (2) to improve the classification accuracy, and (3) to improve the reliability of the estimation of the performance, since a reduced feature set requires less training samples in the training procedure of a pattern classifier (Jain et al., 2000). Feature selection produces savings in the measuring features (since some of the features are discarded) and the selected features retain their original physical

In previous papers (Bouatmane et al., 2007), we addressed the high input dimensionality problem by selecting the best-subset of sequential forward selection SFS followed by a classification using a nearest neighbour classifier (1NN) technique. Although, this approach produced results superior to previously reported methods (Roula, 2002,2003) , the classification accuracy can be further improved by decomposing this multiclass problem into a number of simpler two-class problems. In this case, each subproblem can be regarded separately and solved using a suitable binary classifier. The outputs of this collection of classifiers can then be combined to produce the overall result for the original multiclass problem. In this paper, we propose a Round-Robin (RR) classification algorithm using a sequential forward selection/nearest neighbor (SFS/1NN) classifier to improve the classification accuracy. Round Robin classification is a technique which is suitable for use in multiclass problems. The technique consists of dividing the multiclass problem into an appropriate number of simpler binary classification problems (Furnkranz, 2002). Each binary classifier is implemented as an SFS/1NN classifier, and the final outcome is computed using majority voting technique. A key characteristic of this approach is that, in a binary class, the classifier attempts to find the features that only distinguish that particular class. Thus, different features are selected for each binary classifier, resulting in an overall increase in the classification accuracy. In contrast, in a multiclass problem, the classifier tries

The remainder of this chapter is organized as follows: Sect. 2 gives a description of the dataset used including texture and structural features. Section 3 is concerned with feature selection problem and describes the RR approach followed by the probability estimate for the classifier outputs and error estimation. Sect. 4 describes the image acquisition and dataset. Sect. 5 gives the results obtained and their analysis and discussion including a performance comparative study. Sect. 6 analyses the features selected and sect. 7 gives ROC

Over the last years, the most prolific and promising works in the area of cancer classification have been in the area of texture analysis of the nucleus (Tabesh et al., 2005; Liu et al., 20). This is not surprising since pre-cancerous abnormalities are manifested in visual and subvisual changes in cell characteristics. In fact, it is generally believed that the initial signs of cell neoplasia appear in the nucleus. Because nuclear chromatin and its spatial arrangement can be viewed as a type of texture and whether tissue samples are examined at low, medium or high magnification, texture is a key element in the differentiation between

However, texture features are not sufficient to classify all the groups. The complex structures present in BPH, PIN and also PCa need a higher level description. Thus, structural features, based on segmentation, have been computed for different spectral bands and consolidated in a

large feature vector. The features used are described in the following subsections.

to find those features that distinguish all classes at once.

curves and finally sect. 8 gives a summary of the chapter.

**2. Images features** 

normal and malignant tissue patterns.

interpretation.

To identify prostatic patterns, texture features are needed as a discriminative measurement for the samples. Haralick (Haralick, 1979) assumed that texture information is sufficiently identified by a matrix indexed by grey levels and where the elements represent the frequency of having two defined grey levels separated by a defined distance in a defined direction. This matrix is called grey level co-occurrence matrix (GLCM):

$$\text{cov}(\mathbf{i}\_{\prime}\mathbf{j}\_{\prime}, d\_{\prime}\theta) = \alpha \tag{1}$$

The above equation means that there are α pairs of pixels having i and j respectively, as grey levels and separated by the cylindrical co-ordinate [*d*, ]. The values of d, for which the GLCM is computed, depend on the nature of the texture. Small d values are suitable for fine textures, whereas larger distances are needed to measure coarse textures.

For an image of 256 grey levels (Ng=256), there would be 65536 feature elements to use as a measure for the texture. Therefore, the direct use of the co-occurrence matrix is computationally intensive and as such is not practical. Instead, the texture features are represented by deriving some more meaningful measurements. A set of features was proposed by Haralick to characterise the homogeneity, the coarseness, the periodicity and the linearity of textures. These features are defined as follows:

Angular Second Moment

$$ASM = \sum\_{i=1}^{N\_g} \sum\_{j=1}^{N\_g} P(i, j)^2 \tag{2}$$

Contrast or difference moment

$$\text{CONN} = \sum\_{i=1}^{N\_{\vec{x}}} \sum\_{j=1}^{N\_{\vec{x}}} (i-j)^2 P(i,j) \tag{3}$$

Dissimilarity

$$DSI = \sum\_{i=1}^{N\_{\mathcal{S}}} \sum\_{j=1}^{N\_{\mathcal{S}}} \left| i - j \right| P(i, j) \tag{4}$$

Correlation

$$COR = \sum\_{i=1}^{N\_x} \sum\_{j=1}^{N\_y} \left[ (i - \mu\_x)(j - \mu\_y)P(i, j)^2 \right] / (\sigma\_x \sigma\_y) \tag{5}$$

Where ,,, *x <sup>y</sup> <sup>x</sup> <sup>y</sup>* are the means and the variances of the row sums and column sums of the co-occurrence matrix respectively. Entropy or randomness

$$ENT = -\sum\_{i=1}^{N\_{\mathcal{S}}} \sum\_{j=1}^{N\_{\mathcal{S}}} P(i, j) \log P(i, j) \tag{6}$$

Inverse difference Moment:

$$IDM = -\sum\_{i=1}^{N\_\mathcal{S}} \sum\_{j=1}^{N\_\mathcal{S}} \frac{P(i,j)}{\left(1 + \left(i - j\right)^2\right)}\tag{7}$$

Improving Prostate Cancer Classification:

**3.1 Feature selection problem** 

components for the classification.

reported so far use the wrapper methods.

classification accuracy.

A Round Robin Forward Sequential Selection Approach 135

As discussed in Section 1, the major problem arising from multispectral data is related to the feature vector size. Typically, with 16 bands and 8 features in each band, the feature vector size is 128. For such a high dimensionality problem, pattern recognition techniques suffer from the well-known curse-of-dimensionality problem: keeping the number of training samples limited and increasing the number of features will eventually result in badly

PCA (a well-known unsupervised feature extraction method) has been used by Roula et al. on the large resulting feature vectors to reduce its dimensionality to a manageable size. In their work, Roula et al. used PCA and a linear discrimination function on significant PCA

Another technique to reduce the dimensionality of the feature space is by using feature selection methods. The term feature selection refers to the selection of the best subset of the input feature set. This results in a feature selection producing a smaller set of features (since some of the features are discarded) with the selected features retaining their original physical interpretation. This feature selection problem can be viewed as a multiobjective optimization problem since it involves minimizing the feature subset while maximizing

Mathematically, the feature selection problem can be formulated as follows: Suppose Y is an original feature vector with cardinality n, XY, J(X) is the selection criterion function for the new feature vector X. The goal is to optimize J(X). The choice of an algorithm for selecting the features from an initial set depends on n. The feature selection problem is said to be of small scale, medium scale, or large scale accordingly as n belongs to the intervals [0,19],

Generally, feature selection algorithms have two components: a selection algorithm that generates proposed subsets of features and attempts to find an optimal subset; and an evaluation algorithm which determines how 'good' a proposed feature subset is, by returning some measure of goodness to the selection algorithm. However, without a suitable stopping criterion the feature selection process may run exhaustively or forever through the space of subsets. Stopping criteria can be: (i) whether addition (or deletion) of any feature does not produce a better subset; and (ii) whether an optimal subset according to some evaluation function is obtained. Ideally, a feature selection method searchs through the subsets of features, and tries to find the best one among all the competing candidate subsets according to some evaluation function. However, this procedure is exhaustive as it tries to find only the best one. It may be too costly and practically prohibitive, even for a medium-sized feature set size. Other methods based on heuristic or random search methods; attempt to reduce

[20,49], or [50,+∞], respectively (Duda et al., 2001; Kudo & Sklansky, 2000).

computational complexity by compromising performance (Davies & Russell, 1994).

In (Dash & Liu, 1997) different feature selection methods are categorized into two broad groups (i.e., filter and wrapper) depending on the type of classification algorithm used for the selection of the subset. For example, the filter methods do not require a feedback from the classifier and estimate the classification performance by some indirect assessments, such as distance measures which reflect how well the classes separate from each other. On the other hand, the wrapper methods are classifier-dependent. Based on the classification accuracy, the methods evaluate the "goodness" of the selected feature subset directly, which should intuitively lead to a better performance. Currently, many experimental results

**3. Classification of prostate cancer using round robin approach** 

performing classifiers (Jain et al., 2000; Jimenez, & Landgrebe, 1998).

#### **2.2 Structural features**

The use of texture features alone is not sufficient to capture the complexity of the patterns in prostatic neoplasia. Although, the classification of stroma is relatively simple because of its homogenous nature at low resolution, BPH and PCa present more complex structures, as both can contain glandular areas and nuclei clusters as well. The glandular areas are smaller in regions exhibiting PCa while the nuclear clusters are much larger. The PIN pattern is an intermediate state between the BPH and PCa. It appears that accurate classification requires the quantification of these differences. Segmenting the glandular and the nuclear areas can achieve this quantification, as the glandular areas are lighter compared to the surrounding tissue, while the nuclear clusters are darker (Larsh et al., 2002; Roula et al., 2002).

Figure 2 summarises the segmentation scheme. From the segmented images 1 and 2, two features, f1 and f2 can be computed

$$f\_1 = \bigvee\_{\mathcal{W}^2} \tag{8}$$

$$f\_2 = \bigvee\_{W^2}^{G} \tag{9}$$

Where G and N are the number of pixels segmented as glandular area and classified as nuclear area, respectively. W is the size of the analysis window. These two features allow the quantification of how much nuclear clusters and glandular areas are present in the samples.

Fig. 2. Segmentation of nuclei and glandular areas

#### **3. Classification of prostate cancer using round robin approach**

#### **3.1 Feature selection problem**

134 Prostate Cancer – Original Scientific Reports and Case Studies

The use of texture features alone is not sufficient to capture the complexity of the patterns in prostatic neoplasia. Although, the classification of stroma is relatively simple because of its homogenous nature at low resolution, BPH and PCa present more complex structures, as both can contain glandular areas and nuclei clusters as well. The glandular areas are smaller in regions exhibiting PCa while the nuclear clusters are much larger. The PIN pattern is an intermediate state between the BPH and PCa. It appears that accurate classification requires the quantification of these differences. Segmenting the glandular and the nuclear areas can achieve this quantification, as the glandular areas are lighter compared to the surrounding

Figure 2 summarises the segmentation scheme. From the segmented images 1 and 2, two

<sup>1</sup> <sup>2</sup> *<sup>f</sup> <sup>N</sup>*

<sup>2</sup> <sup>2</sup> *<sup>f</sup> <sup>G</sup>*

Where G and N are the number of pixels segmented as glandular area and classified as nuclear area, respectively. W is the size of the analysis window. These two features allow the quantification of how much nuclear clusters and glandular areas are present in the samples.

Original image

Histogram equalisation

Deriche filtering a=0.9, filter size=3

Threshold 2 at 10% Threshold 1 at 90%

*<sup>W</sup>* (8)

*<sup>W</sup>* (9)

Image 2: segmented glands

tissue, while the nuclear clusters are darker (Larsh et al., 2002; Roula et al., 2002).

**2.2 Structural features** 

features, f1 and f2 can be computed

Fig. 2. Segmentation of nuclei and glandular areas

Image 1: segmented nuclei

As discussed in Section 1, the major problem arising from multispectral data is related to the feature vector size. Typically, with 16 bands and 8 features in each band, the feature vector size is 128. For such a high dimensionality problem, pattern recognition techniques suffer from the well-known curse-of-dimensionality problem: keeping the number of training samples limited and increasing the number of features will eventually result in badly performing classifiers (Jain et al., 2000; Jimenez, & Landgrebe, 1998).

PCA (a well-known unsupervised feature extraction method) has been used by Roula et al. on the large resulting feature vectors to reduce its dimensionality to a manageable size. In their work, Roula et al. used PCA and a linear discrimination function on significant PCA components for the classification.

Another technique to reduce the dimensionality of the feature space is by using feature selection methods. The term feature selection refers to the selection of the best subset of the input feature set. This results in a feature selection producing a smaller set of features (since some of the features are discarded) with the selected features retaining their original physical interpretation. This feature selection problem can be viewed as a multiobjective optimization problem since it involves minimizing the feature subset while maximizing classification accuracy.

Mathematically, the feature selection problem can be formulated as follows: Suppose Y is an original feature vector with cardinality n, XY, J(X) is the selection criterion function for the new feature vector X. The goal is to optimize J(X). The choice of an algorithm for selecting the features from an initial set depends on n. The feature selection problem is said to be of small scale, medium scale, or large scale accordingly as n belongs to the intervals [0,19], [20,49], or [50,+∞], respectively (Duda et al., 2001; Kudo & Sklansky, 2000).

Generally, feature selection algorithms have two components: a selection algorithm that generates proposed subsets of features and attempts to find an optimal subset; and an evaluation algorithm which determines how 'good' a proposed feature subset is, by returning some measure of goodness to the selection algorithm. However, without a suitable stopping criterion the feature selection process may run exhaustively or forever through the space of subsets. Stopping criteria can be: (i) whether addition (or deletion) of any feature does not produce a better subset; and (ii) whether an optimal subset according to some evaluation function is obtained. Ideally, a feature selection method searchs through the subsets of features, and tries to find the best one among all the competing candidate subsets according to some evaluation function. However, this procedure is exhaustive as it tries to find only the best one. It may be too costly and practically prohibitive, even for a medium-sized feature set size. Other methods based on heuristic or random search methods; attempt to reduce computational complexity by compromising performance (Davies & Russell, 1994).

In (Dash & Liu, 1997) different feature selection methods are categorized into two broad groups (i.e., filter and wrapper) depending on the type of classification algorithm used for the selection of the subset. For example, the filter methods do not require a feedback from the classifier and estimate the classification performance by some indirect assessments, such as distance measures which reflect how well the classes separate from each other. On the other hand, the wrapper methods are classifier-dependent. Based on the classification accuracy, the methods evaluate the "goodness" of the selected feature subset directly, which should intuitively lead to a better performance. Currently, many experimental results reported so far use the wrapper methods.

Improving Prostate Cancer Classification:

each classification has to be used.

focused on by c-1 classifiers.

**3.3 Error estimation** 

A Round Robin Forward Sequential Selection Approach 137

The objects are then classified by applying a combination rule on the set of decisions. One strategy is to use voting where the object is labeled to the class with the highest number of votes. When classifying an unknown new sample, each classifier (1NN in this case) determines to which of its two classes the sample is more likely to belong. In this case, we are faced with the possibility of ties. To avoid these ties, a probability estimate value for

In pattern recognition, 1NN is one of the simplest and most widely used algorithms. Given a query sample x, a 1NN algorithm determines the closest neighbor of x in the training nodes using some distance metric (e.g. Euclidean distance in our study) and predicts the class label of the nearest node. In contrast to other statistical classifiers, 1NN needs no model to fit. This property simplifies the structure of the training process by avoiding model training, thus

For the sake of probability estimates, probabilistic outputs of the classifier were required rather than label prediction. For 1NN, objects are assigned to the class of the nearest object in the training set. Posterior probabilities are estimated by comparing the nearest neighbor distances for all classes (Duin & Tax, 1998). A RR ensemble converts a c-class problem into a series of two-class problems by creating one classifier for each pair of classes. New items are classified by submitting them to the c(c-1)/2 binary predictors. The final prediction is achieved by a majority voting. The probability of a query q belonging to a class c can be

(| ) (| )

where M is the set of ensemble members, mc is the class predicted by m and Pm(c|q) is the posterior probability given by ensemble predictor m (the binary classifiers). If m does not involve class c, then Pm(c|q) = 0. The probability estimates of the binary classifiers will be combined using the maximum rule; therefore the instances are assigned to the class with the maximum output given by equation (10). Clearly, RR is a problem decomposition technique. However, there are some aggregation benefits as each class is

Given a small set of samples, appropriate strategies for learning and testing become very critical to avoid over-fitting. Leave-one-out (LOO) and k-fold cross-validation are two popular error estimation procedures to reduce bias in machine learning and testing problems especially with small sample size (sss) (Jain et al., 2000). The procedure of LOO method is to take one out of n observations and use the remaining n-1 observations as the training set for deriving the parameters of the classifier. The classifier is then used to classify the removed observation. This process is repeated for all n observations in order to obtain the estimation of the classification accuracy. In the case of k-fold cross-validation method, the entire sample set is randomly partitioned into k disjoint subsets of equal size, where n is the total number of samples in the entire set. Then, k -1 subsets are used to train the classifier and the remaining subset is used to test for accuracy estimation. This process is repeated for all distinct choices of k subsets and the average of correct classification rates is

*m M*

 

*pcq*

calculated. Notice that k -fold cross-validation is reduced to LOO if k=n.

( ) ( | ).1

(10)

*m mc c*

*p cq*

*m m M*

*p cq*

training with a 1NN classifier only requires selecting the appropriate features.

calculated as follows (Grimaldi et al., 2003) (equation 10):

In this work, an SFS algorithm, which is simple and empirically successful, is proposed for feature selection. It starts with an empty subset of features and performs a hill-climbing deterministic search. At each iteration, a feature not yet selected is individually incorporated in the subset to calculate a criterion. Then the feature which yields the best criterion value is included in the new subset. This iteration will not be stopped until no improvement of the criterion value is achieved. SFS is used as a wrapper approach, therefore the criterion employed to carry out the search is based on error estimation by the selected features using 1NN classifier. In addition, we propose another scheme in which the multiclass problem is addressed using Round Robin (RR) classification approach where the classification problem is decomposed into a number of binary classes. The key point is that it is then possible to design simpler and more efficient binary classifiers as will be demonstrated in the next Section.

#### **3.2 Round robin method**

The RR or pairwise class binarization transforms a c-class problem into c(c − 1)/2 two-class problems i, j with one for each set of classes i, j (i = 1, . . . , c − 1, j = i + 1, . . . , c). A binary classifier for problem i, j is trained with examples of classes i and j, whereas examples of classes k ≠ i, j are ignored for this problem (Furnkranz, 2002). Figure 3 illustrates a multiclass (four-class) learning problem where one classifier (SFS/1NN classifier in this study) separates all classes. Figure 4 shows Round Robin learning with c(c − 1)/2 classifiers. For a four-class problem, the Round Robin trains six classifiers, one for each pair of classes. Each class is trained using a feature selection algorithm based on the SFS/1NN classifier.

Fig. 3. Multiclass learning

Fig. 4. Round Robin learning. p: PIN. c: PCa. b: BPH. s: STR.

In this work, an SFS algorithm, which is simple and empirically successful, is proposed for feature selection. It starts with an empty subset of features and performs a hill-climbing deterministic search. At each iteration, a feature not yet selected is individually incorporated in the subset to calculate a criterion. Then the feature which yields the best criterion value is included in the new subset. This iteration will not be stopped until no improvement of the criterion value is achieved. SFS is used as a wrapper approach, therefore the criterion employed to carry out the search is based on error estimation by the selected features using 1NN classifier. In addition, we propose another scheme in which the multiclass problem is addressed using Round Robin (RR) classification approach where the classification problem is decomposed into a number of binary classes. The key point is that it is then possible to design simpler and more efficient binary classifiers as will be

The RR or pairwise class binarization transforms a c-class problem into c(c − 1)/2 two-class problems i, j with one for each set of classes i, j (i = 1, . . . , c − 1, j = i + 1, . . . , c). A binary classifier for problem i, j is trained with examples of classes i and j, whereas examples of classes k ≠ i, j are ignored for this problem (Furnkranz, 2002). Figure 3 illustrates a multiclass (four-class) learning problem where one classifier (SFS/1NN classifier in this study) separates all classes. Figure 4 shows Round Robin learning with c(c − 1)/2 classifiers. For a four-class problem, the Round Robin trains six classifiers, one for each pair of classes. Each

class is trained using a feature selection algorithm based on the SFS/1NN classifier.

STROMA

BPH

PCa

demonstrated in the next Section.

**3.2 Round robin method** 

Fig. 3. Multiclass learning

Fig. 4. Round Robin learning. p: PIN. c: PCa. b: BPH. s: STR.

PIN

The objects are then classified by applying a combination rule on the set of decisions. One strategy is to use voting where the object is labeled to the class with the highest number of votes. When classifying an unknown new sample, each classifier (1NN in this case) determines to which of its two classes the sample is more likely to belong. In this case, we are faced with the possibility of ties. To avoid these ties, a probability estimate value for each classification has to be used.

In pattern recognition, 1NN is one of the simplest and most widely used algorithms. Given a query sample x, a 1NN algorithm determines the closest neighbor of x in the training nodes using some distance metric (e.g. Euclidean distance in our study) and predicts the class label of the nearest node. In contrast to other statistical classifiers, 1NN needs no model to fit. This property simplifies the structure of the training process by avoiding model training, thus training with a 1NN classifier only requires selecting the appropriate features.

For the sake of probability estimates, probabilistic outputs of the classifier were required rather than label prediction. For 1NN, objects are assigned to the class of the nearest object in the training set. Posterior probabilities are estimated by comparing the nearest neighbor distances for all classes (Duin & Tax, 1998). A RR ensemble converts a c-class problem into a series of two-class problems by creating one classifier for each pair of classes. New items are classified by submitting them to the c(c-1)/2 binary predictors. The final prediction is achieved by a majority voting. The probability of a query q belonging to a class c can be calculated as follows (Grimaldi et al., 2003) (equation 10):

$$p(c|\,q) = \frac{\sum\_{m \in M} p\_m(c|\,q) \cdot \mathbf{1}\_{\{m \gets c\}}}{\sum\_{m \in M} p\_m(c|\,q)} \tag{10}$$

where M is the set of ensemble members, mc is the class predicted by m and Pm(c|q) is the posterior probability given by ensemble predictor m (the binary classifiers). If m does not involve class c, then Pm(c|q) = 0. The probability estimates of the binary classifiers will be combined using the maximum rule; therefore the instances are assigned to the class with the maximum output given by equation (10). Clearly, RR is a problem decomposition technique. However, there are some aggregation benefits as each class is focused on by c-1 classifiers.

#### **3.3 Error estimation**

Given a small set of samples, appropriate strategies for learning and testing become very critical to avoid over-fitting. Leave-one-out (LOO) and k-fold cross-validation are two popular error estimation procedures to reduce bias in machine learning and testing problems especially with small sample size (sss) (Jain et al., 2000). The procedure of LOO method is to take one out of n observations and use the remaining n-1 observations as the training set for deriving the parameters of the classifier. The classifier is then used to classify the removed observation. This process is repeated for all n observations in order to obtain the estimation of the classification accuracy. In the case of k-fold cross-validation method, the entire sample set is randomly partitioned into k disjoint subsets of equal size, where n is the total number of samples in the entire set. Then, k -1 subsets are used to train the classifier and the remaining subset is used to test for accuracy estimation. This process is repeated for all distinct choices of k subsets and the average of correct classification rates is calculated. Notice that k -fold cross-validation is reduced to LOO if k=n.

Improving Prostate Cancer Classification:

and n is the total number of patterns.

*<sup>j</sup> x* is the jth feature of the ith pattern, ,

Fig. 5. Prostatic tissue sample viewed at low and medium magnifications

The assessment of the classification performance has been made using three procedures: 4 fold cross-validation, 10 cross-validation and leave-one-out (LOO) which was applied patient-wise. To obtain a k-fold cross-validation estimate of the classification performance, the dataset was randomly split into k sets of a roughly equal size. Splitting was carried out such that the proportion of samples per class was roughly equal across the sets. Each run of the k-fold cross-validation algorithm consisted of a classifier design on k-1 dataset subsets (training) while testing was performed on the remaining subset. The optimal feature subset for each cross-validation run was determined as the subset with the highest LOO accuracy

**5. Experiments and discussion** 

estimate on the corresponding training set.

where *<sup>i</sup>*

A Round Robin Forward Sequential Selection Approach 139

The data were taken from a total of 10 different patients with typically 3-6 biopsies per patient (from different areas in the prostate) and 8-12 images were taken from each samples (from different areas in the image). The dataset consists of textured multispectral images taken at 16 spectral channels (from 500 to 650 nm) (Roula et al., 2002). Five hundred and ninety-two different samples (multispectral images) of size 128 × 128 have been used to carry out the analysis. The samples are examined at low power (40 x objective magnifications) by the two highly experienced independent pathologists and labelled into four classes: 165 cases of Stroma, 106 cases of BPH, 144 cases of PIN, and 177 cases of PCa.

*t*

*<sup>i</sup> <sup>j</sup> x* is the corresponding normalized feature

When referring to the performance of a classification model, we are interested in the model's ability to correctly predict or separate the classes. When looking at the errors made by a classification model, the confusion matrix used in this paper gives the full picture. The confusion matrix shows how accurate the predictions are made by the model. The rows correspond to the known class of the data, i.e. the labels in the data while the columns correspond to the predictions made by the model. The value of each of element in the matrix is the number of predictions made with the class corresponding to the column, for example, with the correct value as represented by the row. Thus, the diagonal elements show the number of correct classifications made for each class, and the off-diagonal elements show the errors made.

Accuracy is the overall correctness of the model and is calculated as the sum of the correct classifications divided by the total number of classifications. Precision is a measure of the accuracy provided that a specific class has been predicted. It is defined by:

$$Precision = \frac{tp}{tp + fp} \tag{11}$$

where tp and fp are the numbers of true positive and false positive predictions for the considered class. Recall is a measure of the ability of a prediction model to select instances of a certain class from a data set. It is also commonly called sensitivity, and corresponds to the true positive rate and can be written as:

$$Recall = Sensitivity = \frac{tp}{tp + fn} \tag{12}$$

where tp and fn are the numbers of true positive and false negative predictions for the considered class. tp+ fn is the total number of test examples of the considered class.

#### **4. Sample preparation, image acquisition and datasets description**

Entire tissue samples were taken from prostate glands. Sections 5-µm thick were extracted and stained using the widely used H&E stains. These samples were routinely assessed by two experienced pathologists and graded histologically as showing STR, BPH, PIN, and PCa. From these samples, whole subimage sections were captured using a classical microscope and CCD camera. An LCTF (VARISPECTM) was inserted in the optical path between the light source and a CCD camera. The LCTF has a bandwidth accuracy of 5 nm. The wavelength is controllable through the visible spectrum (from 400 to 720 nm). This allowed for the capture of multispectral images of the tissue samples by using different spectral frequencies. Figure 5 shows a prostatic tissue sample viewed at different magnification.

In order to offset any bias due to the different range of values for the original features, the input feature values are normalized over the range [1,11] using equation (13) ( Raymer et al., 2000). Normalizing the data is important to ensure that the distance measure allocates equal weight to each variable. Without normalization, the variable with the largest scale will dominate the measure:

$$\mathbf{x}'\_{i,j} = \left(\frac{\mathbf{x}\_{i,j} - \min\_{k=1\ldots n} \mathbf{x}(k,j)}{\max\_{k=1\ldots n} \mathbf{x}(k,j) - \min\_{k=1\ldots n} \mathbf{x}(k,j)} \times 10\right) + 1\tag{13}$$

When referring to the performance of a classification model, we are interested in the model's ability to correctly predict or separate the classes. When looking at the errors made by a classification model, the confusion matrix used in this paper gives the full picture. The confusion matrix shows how accurate the predictions are made by the model. The rows correspond to the known class of the data, i.e. the labels in the data while the columns correspond to the predictions made by the model. The value of each of element in the matrix is the number of predictions made with the class corresponding to the column, for example, with the correct value as represented by the row. Thus, the diagonal elements show the number of correct classifications made for each class, and the off-diagonal elements show

Accuracy is the overall correctness of the model and is calculated as the sum of the correct classifications divided by the total number of classifications. Precision is a measure of the

*tp Precision*

where tp and fp are the numbers of true positive and false positive predictions for the considered class. Recall is a measure of the ability of a prediction model to select instances of a certain class from a data set. It is also commonly called sensitivity, and corresponds to the

*tp Recall Sensitivity tp fn*

where tp and fn are the numbers of true positive and false negative predictions for the

Entire tissue samples were taken from prostate glands. Sections 5-µm thick were extracted and stained using the widely used H&E stains. These samples were routinely assessed by two experienced pathologists and graded histologically as showing STR, BPH, PIN, and PCa. From these samples, whole subimage sections were captured using a classical microscope and CCD camera. An LCTF (VARISPECTM) was inserted in the optical path between the light source and a CCD camera. The LCTF has a bandwidth accuracy of 5 nm. The wavelength is controllable through the visible spectrum (from 400 to 720 nm). This allowed for the capture of multispectral images of the tissue samples by using different spectral frequencies. Figure 5 shows a prostatic tissue sample viewed at different

In order to offset any bias due to the different range of values for the original features, the input feature values are normalized over the range [1,11] using equation (13) ( Raymer et al., 2000). Normalizing the data is important to ensure that the distance measure allocates equal weight to each variable. Without normalization, the variable with the largest scale will

, 1...

1... 1...

*k n k n x x k j*

max ( , ) min ( , ) *ij k n*

 

*xk j xk j*

min ( , ) 10 1

considered class. tp+ fn is the total number of test examples of the considered class.

**4. Sample preparation, image acquisition and datasets description** 

*tp fp*

(11)

(12)

(13)

accuracy provided that a specific class has been predicted. It is defined by:

the errors made.

magnification.

dominate the measure:

,

*x*

*i j*

true positive rate and can be written as:

where *<sup>i</sup> <sup>j</sup> x* is the jth feature of the ith pattern, , *t <sup>i</sup> <sup>j</sup> x* is the corresponding normalized feature and n is the total number of patterns.

The data were taken from a total of 10 different patients with typically 3-6 biopsies per patient (from different areas in the prostate) and 8-12 images were taken from each samples (from different areas in the image). The dataset consists of textured multispectral images taken at 16 spectral channels (from 500 to 650 nm) (Roula et al., 2002). Five hundred and ninety-two different samples (multispectral images) of size 128 × 128 have been used to carry out the analysis. The samples are examined at low power (40 x objective magnifications) by the two highly experienced independent pathologists and labelled into four classes: 165 cases of Stroma, 106 cases of BPH, 144 cases of PIN, and 177 cases of PCa.

Fig. 5. Prostatic tissue sample viewed at low and medium magnifications

#### **5. Experiments and discussion**

The assessment of the classification performance has been made using three procedures: 4 fold cross-validation, 10 cross-validation and leave-one-out (LOO) which was applied patient-wise. To obtain a k-fold cross-validation estimate of the classification performance, the dataset was randomly split into k sets of a roughly equal size. Splitting was carried out such that the proportion of samples per class was roughly equal across the sets. Each run of the k-fold cross-validation algorithm consisted of a classifier design on k-1 dataset subsets (training) while testing was performed on the remaining subset. The optimal feature subset for each cross-validation run was determined as the subset with the highest LOO accuracy estimate on the corresponding training set.

Improving Prostate Cancer Classification:

1996) on a wide range of datasets.

0

Round robin SFS/1NN

Round robin SFS/1NN

Table 1. Comparison of error classification rate

0.2

0.4

0.6

**fraction of selection**

0.8

1

1.2

A Round Robin Forward Sequential Selection Approach 141

Bagging is a general method of combining classifiers that can be applied to any base method. It is a relatively simple idea: n datasets are created by sampling the patterns with replacement from the original training set. Each of the n datasets has the same number of patterns as the original training set. A classifier is then trained on each dataset by combining the outputs using simple voting. Bagging has obtained impressive error reductions with decision trees such as CART (Breiman, 1996) and C4.5 (Freund & Schapire, 1996; Quinlan,

> 1 9 17 25 33 41 49 57 65 73 81 89 97 105 113 121 **features**

> > 10 cross validation error estimation %

Leave-one-out error estimation %

Fig. 7. Subsets yielded by application of the SFS from 4 cross-validations

4 cross-validation error estimation %

1NN classifier 13.34 12.18 12

SFS/1NN classifier 10.22 7.41 3.37

(voting rule) 10.98 9.62 2.87

(maximum probability rule) 8.91 7.26 0.17

In Boosting, the classifiers in the ensemble are trained serially, with the weights on the training instances set adaptively according to the performance of the previous classifiers. If

The first aim was to determine the optimum number of features to obtain the best achievable classification performance. Therefore, the feature selection algorithm SFS described in section 3 with a 1NN classifier was used. Figure 6 shows the results obtained using LOO error estimation. The curve representing the results from the feature selection shows a strong increase in performance for small subsets followed by slight increase up to medium sized subsets. Large subsets cause a drop in the recognition rate.

For k-fold cross-validation the results show that using SFS with different training sets does not yield identical feature subsets. This is illustrated by the diagram in Figure 7 which shows the fraction of how often a feature was selected divided by the total number of simulations using 4 cross-validation method. One can see that the selected features originate from different spectral bands.

#### Fig. 6. Recognition rate of SFS algorithm

The accuracies of the selected features subsets are given in Table 1. The combination of the binary classifiers' results generated by proposed Round Robin algorithm is performed using two methods: the voting rule using the resulted classes (Tahir & Bouridane, 2006) and the maximum probability obtained using equation (10). For all the cross validation estimations, the RR SFS/1NN with the maximum probability gives the best classification accuracy. As shown in Table 1, RRSFS algorithm using LOO error estimation achieves the lowest error rate. The overall classification error has been reduced from 3.37% to 0.17%. To gain an insight into the classification of different classes of prostate cancer, the confusion matrix of the multiclass SFS/1NN and the proposed Round Robin learning using SFS/1NN are also given. Table 2 depicts the results using the LOO error estimation where Table 3 gives the corresponding results using 4 cross-validations. Note that in all the cases, BPH and PIN classes present the highest error rate in terms of classification but the use of Round Robin algorithm reduces significantly the error rate in these classes.

The first aim was to determine the optimum number of features to obtain the best achievable classification performance. Therefore, the feature selection algorithm SFS described in section 3 with a 1NN classifier was used. Figure 6 shows the results obtained using LOO error estimation. The curve representing the results from the feature selection shows a strong increase in performance for small subsets followed by slight increase up to

For k-fold cross-validation the results show that using SFS with different training sets does not yield identical feature subsets. This is illustrated by the diagram in Figure 7 which shows the fraction of how often a feature was selected divided by the total number of simulations using 4 cross-validation method. One can see that the selected features originate

The accuracies of the selected features subsets are given in Table 1. The combination of the binary classifiers' results generated by proposed Round Robin algorithm is performed using two methods: the voting rule using the resulted classes (Tahir & Bouridane, 2006) and the maximum probability obtained using equation (10). For all the cross validation estimations, the RR SFS/1NN with the maximum probability gives the best classification accuracy. As shown in Table 1, RRSFS algorithm using LOO error estimation achieves the lowest error rate. The overall classification error has been reduced from 3.37% to 0.17%. To gain an insight into the classification of different classes of prostate cancer, the confusion matrix of the multiclass SFS/1NN and the proposed Round Robin learning using SFS/1NN are also given. Table 2 depicts the results using the LOO error estimation where Table 3 gives the corresponding results using 4 cross-validations. Note that in all the cases, BPH and PIN classes present the highest error rate in terms of classification but the use of Round Robin

1 11 21 31 41 51 61 71 81 91 101 111 121

**Number of selected features**

medium sized subsets. Large subsets cause a drop in the recognition rate.

from different spectral bands.

**Classification rate**

Fig. 6. Recognition rate of SFS algorithm

0,40

0,50

0,60

0,70

0,80

0,90

1,00

algorithm reduces significantly the error rate in these classes.

Bagging is a general method of combining classifiers that can be applied to any base method. It is a relatively simple idea: n datasets are created by sampling the patterns with replacement from the original training set. Each of the n datasets has the same number of patterns as the original training set. A classifier is then trained on each dataset by combining the outputs using simple voting. Bagging has obtained impressive error reductions with decision trees such as CART (Breiman, 1996) and C4.5 (Freund & Schapire, 1996; Quinlan, 1996) on a wide range of datasets.

Fig. 7. Subsets yielded by application of the SFS from 4 cross-validations


Table 1. Comparison of error classification rate

In Boosting, the classifiers in the ensemble are trained serially, with the weights on the training instances set adaptively according to the performance of the previous classifiers. If

Improving Prostate Cancer Classification:

only those features that distinguish its own binary classes.

Feature selection method

SFS/1NN

SFS/1NN Round Robin

Table 5. Number of Features Used By Different Classifiers

bagging and boosting.

and PCa.

A Round Robin Forward Sequential Selection Approach 143

Table 4 shows the comparison between the RR-SFS/1NN versus Bagging and AdaBoost. Decision Tree (C4.5) ( Quinlan, 1993) and NN classifiers are used as base classifiers for

Unfortunately, bagging and boosting are unable to improve the classification accuracy when an NN classifier is used as a base classifier (Yongguang et al., 2004). This fact is clearly seen from Table 6 where the classification accuracy is degraded while using AdaBoost, and only minor improvements are achieved when using bagging. However, the classification accuracy is improved by using bagging and boosting when C4.5 is used as base classifier. Furthermore, it is clear from the table that the proposed Round Robin ensemble technique using TS/1NN has outperformed both bagging and boosting ensemble-design techniques. A key characteristic of the proposed Round Robin approach is that different features are captured and used for each binary classifier in the four-class problem, thus producing an overall increase in the classification accuracy. In contrast, in a multiclass problem, the classifier tries to find those features that distinguish all four-classes at once. Furthermore, the inherent curse-of-dimensionality problem, which arises in a multispectral data, is also resolved by the RR SFS/1NN classifiers since each classifier is trained to compute and use

Table 5 shows the number of features used by the ensemble of binary classifiers. Different numbers of features have been used by the various binary classifiers producing an overall increase in the classification accuracy. Fc represents those features that are common in two or more different binary classifiers. The total number of features in the proposed Round Robin technique is comparable with the multiclass SFS/1NN with lower error rate, but the number of features used by each binary classifier is smaller than that used in other methods. Consequently, multispectral data is better utilized by using a Round Robin technique since the use of more features means more information is captured and used in the classification process. Furthermore, simple binary classes are also useful for analyzing features and are extremely helpful for pathologists in distinguishing various patterns such as BPH, PIN, STR,

SFS/1NN Multi-class 13

Binary-class (stroma-Bph) Binary-class (Stroma-Pin) Binary-class (stroma-PCa) Binary-class (Bph-Pin) Binary-class (Bph-PCa) Binary-class (Pin-PCa)

Features used

6

1

*F F* 

*i*

(15 3) 12 *<sup>c</sup>*

SFS/1NN multiclass learning Round Robin SFS/1NN learning Classified as: BPH PCa PIN Stroma Error (%) BPH PCa PIN Stroma Error (%) BPH 101 0 0 5 4.71 106 0 0 0 0 PCa 1 174 2 0 1.69 0 177 0 0 0 PIN 0 2 137 5 4.86 0 0 143 1 0.69 Stroma 5 0 0 160 3.03 0 0 0 165 0 overall 3.37 0.17

the classifier does not directly support weighted instances, this can be simulated by sampling from the training set with a probability proportional to an instance weight. The main idea is that the classification algorithm should concentrate on the difficult instances.

Table 2. Classification Error by multiclass and round robin learning using SFS/1NN and loo error estimation


Table 3. Classification error by multiclass and round robin learning using SFS/1NN and 4cross-validation


Table 4. Classification accuracy (%) using various ensemble techniques

the classifier does not directly support weighted instances, this can be simulated by sampling from the training set with a probability proportional to an instance weight. The main idea is that the classification algorithm should concentrate on the difficult instances.

SFS/1NN multiclass learning Round Robin SFS/1NN learning

BPH 101 0 0 5 4.71 106 0 0 0 0

PCa 1 174 2 0 1.69 0 177 0 0 0

PIN 0 2 137 5 4.86 0 0 143 1 0.69

Stroma 5 0 0 160 3.03 0 0 0 165 0

overall 3.37 0.17

Table 2. Classification Error by multiclass and round robin learning using SFS/1NN and loo

Round Robin SFS/1NN learning SFS/1NN multiclass learning

BPH 96 0 0 10 9.43 93 3 2 8 12.26

PCa 1 164 8 4 7.43 2 163 11 1 7.90

PIN 0 13 129 2 10.41 3 8 122 11 15.27

Stroma 8 1 5 151 8.48 5 1 3 156 5.45

overall 8.91 10.22

C4.5 Nearest neighbor

C4.5 Bagging Boosting NN Bagging Boosting RR-

91.6 93.2 95.4 88.0 89.2 88.1 99.83

Table 4. Classification accuracy (%) using various ensemble techniques

Table 3. Classification error by multiclass and round robin learning using SFS/1NN and

(%) BPH PCa PIN Stroma Error

(%) BPH PCa PIN Stroma Error

(%)

(%)

SFS

BPH PCa PIN Stroma Error

BPH PCa PIN Stroma Error

Classified as:

error estimation

4cross-validation

Classified as:

Table 4 shows the comparison between the RR-SFS/1NN versus Bagging and AdaBoost. Decision Tree (C4.5) ( Quinlan, 1993) and NN classifiers are used as base classifiers for bagging and boosting.

Unfortunately, bagging and boosting are unable to improve the classification accuracy when an NN classifier is used as a base classifier (Yongguang et al., 2004). This fact is clearly seen from Table 6 where the classification accuracy is degraded while using AdaBoost, and only minor improvements are achieved when using bagging. However, the classification accuracy is improved by using bagging and boosting when C4.5 is used as base classifier. Furthermore, it is clear from the table that the proposed Round Robin ensemble technique using TS/1NN has outperformed both bagging and boosting ensemble-design techniques.

A key characteristic of the proposed Round Robin approach is that different features are captured and used for each binary classifier in the four-class problem, thus producing an overall increase in the classification accuracy. In contrast, in a multiclass problem, the classifier tries to find those features that distinguish all four-classes at once. Furthermore, the inherent curse-of-dimensionality problem, which arises in a multispectral data, is also resolved by the RR SFS/1NN classifiers since each classifier is trained to compute and use only those features that distinguish its own binary classes.

Table 5 shows the number of features used by the ensemble of binary classifiers. Different numbers of features have been used by the various binary classifiers producing an overall increase in the classification accuracy. Fc represents those features that are common in two or more different binary classifiers. The total number of features in the proposed Round Robin technique is comparable with the multiclass SFS/1NN with lower error rate, but the number of features used by each binary classifier is smaller than that used in other methods. Consequently, multispectral data is better utilized by using a Round Robin technique since the use of more features means more information is captured and used in the classification process. Furthermore, simple binary classes are also useful for analyzing features and are extremely helpful for pathologists in distinguishing various patterns such as BPH, PIN, STR, and PCa.


Table 5. Number of Features Used By Different Classifiers

Improving Prostate Cancer Classification:

making such statements?

A Round Robin Forward Sequential Selection Approach 145

real physical dissimilarity between the two groups or is it due to those specific samples? And in the case where it is a real physical difference, what is the level of confidence when

In this work, Student t-test (Montgomery*,* 1997) was used as a statistical test of significance for mean difference of each class pair for all the selected features. Table 6 and 7 show the selected features for SFS/1NN classifier and RR SFS/1NN method using LOO, respectively. The asterisk (\*) in the table shows that this feature exhibits a significant difference in means for all group pairs of cancer (Stroma, BPH, PIN, PCa) with 95 % confidence (p-value<0.05) while (\*\*) shows confidence in difference of means higher than 99% (p-value<0.01). For the

In Table 6, for 9 out of the 13 features selected, the p-value exhibits values lower than 0.05, i.e. yields confidence levels in difference between groups >95%. Three features exhibit a

It was observed that dissimilarity, inverse difference moment, entropy and contrast are the texture features selected. They are all measures of homogeneity of grey level texture. This indicates that prostatic tissues display a clear visual difference in terms of texture. Consequently, neighboring pixels were more likely to have larger grey level differences for different grades of malignancy. Note that the features which have not asterisks exhibit

Rank Selected features Spectral band

1 Inverse difference moment \* 13 2 structural (f2) \* 9 3 Structural (f1) 8 4 Dissimilarity \* 3 5 Contrast 7 6 Structural (f1) \*\* 7 7 Structural (f2) 5 8 Dissimilarity 11 9 Contrast \* 4 10 Entropy \*\* 8 11 Contrast \*\* 6 12 Inverse difference moment 8 13 Structural (f1) \* 11

round robin method, t-test was run only for the binary classes.

significant difference in means, but not for all the pairs of classes.

confidence level in the mean difference superior to 99%.

Table 6. Selected features by SFS/1NN Classifier

Figure 8 shows the results of Recall and Precision measures for different algorithms including the results of Round Robin tabu search RR TS/1NN (Tahir & Bouridane, 2006). From the graphs presented one can observe that for both Precision and Recall, the values of RR SFS/1NN are very high for different classes of prostate cancer. In addition, one can notice from equations (11) and (12) that the values for FP and FN tend to zero when the Precision and Recall tend to 100%. Thus, the false positives and especially false negatives are almost null with our approach. This clearly demonstrates the efficiency of our proposed RR technique.

Fig. 8. Precision and recall measures of classification

#### **6. Analysis of the selected features**

Very often, it is interesting to know if the difference in the mean values for a given feature between two groups is accidental or due to an inherent difference between the groups regarding a specific feature. For example, the mean of a given image feature can be numerically different for normal and cancer prostate cells. But does this difference reflect a

Figure 8 shows the results of Recall and Precision measures for different algorithms including the results of Round Robin tabu search RR TS/1NN (Tahir & Bouridane, 2006). From the graphs presented one can observe that for both Precision and Recall, the values of RR SFS/1NN are very high for different classes of prostate cancer. In addition, one can notice from equations (11) and (12) that the values for FP and FN tend to zero when the Precision and Recall tend to 100%. Thus, the false positives and especially false negatives are almost null with our approach. This clearly demonstrates the efficiency of our

> BPH PCA PIN STROMA **prostate cancer classes**

BPH PCA PIN STROMA **prosate cancer classes**

Very often, it is interesting to know if the difference in the mean values for a given feature between two groups is accidental or due to an inherent difference between the groups regarding a specific feature. For example, the mean of a given image feature can be numerically different for normal and cancer prostate cells. But does this difference reflect a

SFS/NN RR SFS/NN RR TS/NN

> SFS/NN RR SFS/NN RR TS/NN

proposed RR technique.

0,92 0,93 0,94 0,95 0,96 0,97 0,98 0,99 1 1,01

> 0,91 0,92 0,93 0,94 0,95 0,96 0,97 0,98 0,99 1 1,01

Fig. 8. Precision and recall measures of classification

**6. Analysis of the selected features** 

**recall measure of classification**

**preecision measure of** 

**classifiaction**

real physical dissimilarity between the two groups or is it due to those specific samples? And in the case where it is a real physical difference, what is the level of confidence when making such statements?

In this work, Student t-test (Montgomery*,* 1997) was used as a statistical test of significance for mean difference of each class pair for all the selected features. Table 6 and 7 show the selected features for SFS/1NN classifier and RR SFS/1NN method using LOO, respectively. The asterisk (\*) in the table shows that this feature exhibits a significant difference in means for all group pairs of cancer (Stroma, BPH, PIN, PCa) with 95 % confidence (p-value<0.05) while (\*\*) shows confidence in difference of means higher than 99% (p-value<0.01). For the round robin method, t-test was run only for the binary classes.

In Table 6, for 9 out of the 13 features selected, the p-value exhibits values lower than 0.05, i.e. yields confidence levels in difference between groups >95%. Three features exhibit a confidence level in the mean difference superior to 99%.

It was observed that dissimilarity, inverse difference moment, entropy and contrast are the texture features selected. They are all measures of homogeneity of grey level texture. This indicates that prostatic tissues display a clear visual difference in terms of texture. Consequently, neighboring pixels were more likely to have larger grey level differences for different grades of malignancy. Note that the features which have not asterisks exhibit significant difference in means, but not for all the pairs of classes.


Table 6. Selected features by SFS/1NN Classifier

Improving Prostate Cancer Classification:

total number of negative cases (Fawcett, 2003).

0

Fig. 9. ROC curves for SFS/1NN and RR-SFS/1NN classifier

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

True Positive Rate

which are most important for early detection of cancer.

while Pca and PIN form the positive diagnosis outcome.

**7. ROC curve** 

A Round Robin Forward Sequential Selection Approach 147

ROC curve (receiving operating characteristic) analysis has been widely used as a method for medical decisions making. It is a plot of false positive rate (X-axis) versus true positive rate (Y-axis) of a binary classifier. ROC is commonly used for visualizing and selecting classifiers based on their performance. The true positive rate (TPR) is defined as the ratio of the number of correctly classified positive cases to the total number of positive cases. The false positive rate (FPR) is defined as the ratio of incorrectly classified negative cases to the

ROC curves help researchers focus on classification rules with low false positive rates,

The diagonal line y = x corresponds to a classifier which predicts a class membership by randomly guessing it. Hence, all useful classifiers must have ROC curves above this line. We assume that one of the classes is the class of interest and the objects labeled in this class will be called 'positive'. This achieved by considering the BPH as the negative diagnosis

The classifier gives a continuous valued output given by equation (10) which is cut at a certain threshold. All objects for which the classifier output exceeds the threshold are labeled as positive while the remaining results are labeled as negative. By varying the threshold value from the minimum to the maximum value of the classifier output, one can construct a ROC curve. Figure 9 illustrates the ROC curves obtained with the two methods RRSFS/1NN and SFS/1NN using 4 cross-validation and an independent test set. The test set is obtained by splitting the dataset onto two equal sets, training set and test set. For crossvalidation, given the test sets generated from 4 cross-validation, we can simply merge the instances together by their assigned scores into one large test set and we then plot the result.

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

SFS/1NN 4 cross-validation RR SFS/1NN 4 cross-validation

 SFS/1NN test set RR SFS/1NN test set

False Positive Rate


Table 7. Selected features By RR-SFS/1NN Classifier

For the RR SFS/1NN method, all the features presented in Table 7 exhibit a confidence level in mean difference superior to 99% for all the binary classes. This can be explained by the fact that the RRSFS/1NN method selects the features that distinguish only that class. In contrast, in multiclass SFS/1NN, the classifier tries to find those features that distinguish all classes at once.

The presence of structural features can be observed, especially to discriminate BPH and PCa from the other classes. This is because BPH is first characterised by a conspicuous glandular presence. For the PCa, this is due to the predominance of nuclei clusters and the total absence of glands, which makes it easy to detect using the structural features. We note also that the texture features selected are all measures of homogeneity. Contrast is selected alone by Stroma-Pin classifier without the structural features since the glands are totally absent from Stroma. PIN, which is an intermediate state between PCa and BPH may or may not contain lumen glands. Correlation is totally absent from the two tables thus indicating that correlation is a poor discriminant feature. It can be concluded that the joint use of texture and structural features is an efficient method to classify all groups together.

Finally, it is important to see the impact of the multispectral dimension on the classification; the features selected in both methods are from different bands. This shows that the satisfactory results obtained previously are not only due to the adequate choice of features but to the contribution of the multispectral information which characterizes the different classes.

#### **7. ROC curve**

146 Prostate Cancer – Original Scientific Reports and Case Studies

(Stroma-Pin) classifier Contrast \*\* 15

(Bph-Pin) classifier Structural (f1)\*\* 14 (Bph-PCa) classifier Structural (f1)\*\* 14

For the RR SFS/1NN method, all the features presented in Table 7 exhibit a confidence level in mean difference superior to 99% for all the binary classes. This can be explained by the fact that the RRSFS/1NN method selects the features that distinguish only that class. In contrast, in multiclass SFS/1NN, the classifier tries to find those features that distinguish all

The presence of structural features can be observed, especially to discriminate BPH and PCa from the other classes. This is because BPH is first characterised by a conspicuous glandular presence. For the PCa, this is due to the predominance of nuclei clusters and the total absence of glands, which makes it easy to detect using the structural features. We note also that the texture features selected are all measures of homogeneity. Contrast is selected alone by Stroma-Pin classifier without the structural features since the glands are totally absent from Stroma. PIN, which is an intermediate state between PCa and BPH may or may not contain lumen glands. Correlation is totally absent from the two tables thus indicating that correlation is a poor discriminant feature. It can be concluded that the joint use of texture

Finally, it is important to see the impact of the multispectral dimension on the classification; the features selected in both methods are from different bands. This shows that the satisfactory results obtained previously are not only due to the adequate choice of features but to the

contribution of the multispectral information which characterizes the different classes.

and structural features is an efficient method to classify all groups together.

(Stroma-Bph) classifier

(Stroma-PCa) classifier

(Pin-PCa) classifier

classes at once.

Table 7. Selected features By RR-SFS/1NN Classifier

Binary classifiers Selected features Spectral band

Structural (f2)\*\* 11 Dissimilarity \*\* 15 Dissimilarity \*\* 13 Angular second moment \*\* 13

Structural (f2) \*\* 9 Structural (f1)\*\* 3 Inverse difference moment \*\* 5 Dissimilarity \*\* 4

Inverse difference moment \*\* 10 Contrast \*\* 2 Dissimilarity \*\* 4 Structural (f2)\*\* 9

ROC curve (receiving operating characteristic) analysis has been widely used as a method for medical decisions making. It is a plot of false positive rate (X-axis) versus true positive rate (Y-axis) of a binary classifier. ROC is commonly used for visualizing and selecting classifiers based on their performance. The true positive rate (TPR) is defined as the ratio of the number of correctly classified positive cases to the total number of positive cases. The false positive rate (FPR) is defined as the ratio of incorrectly classified negative cases to the total number of negative cases (Fawcett, 2003).

ROC curves help researchers focus on classification rules with low false positive rates, which are most important for early detection of cancer.

The diagonal line y = x corresponds to a classifier which predicts a class membership by randomly guessing it. Hence, all useful classifiers must have ROC curves above this line.

We assume that one of the classes is the class of interest and the objects labeled in this class will be called 'positive'. This achieved by considering the BPH as the negative diagnosis while Pca and PIN form the positive diagnosis outcome.

The classifier gives a continuous valued output given by equation (10) which is cut at a certain threshold. All objects for which the classifier output exceeds the threshold are labeled as positive while the remaining results are labeled as negative. By varying the threshold value from the minimum to the maximum value of the classifier output, one can construct a ROC curve. Figure 9 illustrates the ROC curves obtained with the two methods RRSFS/1NN and SFS/1NN using 4 cross-validation and an independent test set. The test set is obtained by splitting the dataset onto two equal sets, training set and test set. For crossvalidation, given the test sets generated from 4 cross-validation, we can simply merge the instances together by their assigned scores into one large test set and we then plot the result.

Fig. 9. ROC curves for SFS/1NN and RR-SFS/1NN classifier

Improving Prostate Cancer Classification:

67, pp.786-804.

No. 4, pp 1–16.

A Round Robin Forward Sequential Selection Approach 149

Freund, Y. & Schapire, R. E. (1996). Experiments with a new boosting algorithm. Machine Learning, Proceedings *of the Thirteenth International Conference*, pp. 325-332. Furnkranz, J. (2002). Round robin classification*. J. Mach. Learn. Res*., vol. 2,pp. 721–747. Gleason, D. F. & Tannenbaum, M. (1977). The veteran's administration cooperative urologic

*Urologic Pathology: The Prostate*. Philadephia, PA: Lea Febiger, , pp. 171–198. Grimaldi, M.; Cunningham, P. & Kokaram, A. (2003). An evaluation of alternative feature

Haralick, R. M. (1979). Statistical and structural approaches to texture. *Proc. Of the IEEE,* vol.

Huang, P.W. & Lee, C.H. (2009) automatic classification for pathological prostate images based

Jain, A. K.; Duin, R. P. W. & Mao, J*. (2000). Statistical pattern recognition: A review. IEEE Trans.* 

Jimenez, L. O. & Landgrebe, D. A. (1998). Supervised classification in high dimensional

Larsh, P.; Cheriboga, L. Yee, H. & Diem, M. (2002). Infrared spectroscopy of humans cells and tissue: Detection of disease. *Technol*. *Cancer Res. Treat*., vol. 1, no. 1, pp. 1–7. Liu, Y.; Zahoa, T. & Zhang, J. (2002) Learning multispectral texture features for cervical cancer detection. *Proc IEEE Int. Symp. Biomed. Imaging*, Washington, DC, pp. 169–172. Masood, K.& Rajpoot N. (2008). Colon biopsy classification Annals of the BMVA Vol. 2008,

O'Dowd, G. J.; Veltri, R. W.; Miller, M. C. & Strum, S. B.( 2001). The Gleason score: A

Quinlan, J. R. (1996). Bagging, Boosting, and C4.5. *Proceedings of the Thirteenth National* 

Quinlan, R. (1993). *C4.5: Programs for Machine Learning*. San Mateo, CA: Morgan Kaufmann,. Raymer, M. L. et al. (2000)Dimensionality reduction using genetic algorithms. IEEE Trans.

Roula, M. A.; Diamond, J.; Bouridane, A.; Miller, P. & Amira, A. (2002). A multispectral

Roula, M.; A. Bouridane, A. & Miller, P. (2003) A quadratic classifier based on multispectral

Stewart, B. W. & Kleihues, P. (2003). World Cancer Report World Health Organization,

significant biologic manifestation of prostate cancer aggressiveness on biopsy,

computer vision system for automatic grading of prostatic neoplasia. *Proc. IEEE Int.* 

texture features for prostate cancer diagnosis. *Proc. 7th Int. Symp. Signal Process.* 

*IEEE Trans. Syst., Man, Cybern. C, Appl. Rev*., vol. 28, no. 1, pp. 39–54. Kronz, J. D; Westra, W. H & Epstein, J. I. (1999). Mandatory second opinion Surgical Pathology at a Large Referral Hospital, Cancer, vol. 86, no. 11, pp. 2426-2435. Kudo, M. & Sklansky, J. (2000). Comparison of algorithms that select features for pattern

Montgomery*,* D. (1997). *Design and analysis of experiments*. John Wiley & Son, 4th Ed.

*Prostate Cancer Res. Inst.: PCR Insights*, vol. 4, no. 1, pp. 1–5.

*Multimedia Discovery and Mining* at ECML/PKDD.

*Pattern Anal. Mach. Intell*., vol. 22, no. 1, pp. 4–37.

classifiers. *Pattern Recognit*., vol. 33, pp. 25–41.

*Conference on Arti\_cial Intelligence*, 725-730.

Evol. Comput., vol. 4, no. 2, pp. 164–171.

*International Agency for Research on Cancer.* 

*Symp. Biomed. lmaging* , pp. 193–196.

*Appl*., Paris, France, pp. 37–40.

research group: Histologic grading and clinical staging of prostatic carcinoma, in

selection strategies and ensemble techniques of classifying music. *Workshop in* 

on fractal analysis, *IEEE transactions on medical imaging*, VOL. 28, NO. 7, pp.1037–1050.

space: Geometrical, statistical, and asymptotical properties of multivariate data.

The results are comparable or better than those obtained in other recent studies (Taher & Bouridane, 2006); this further demonstrates that our new proposed Round Robin technique results in an improved ability to distinguish cancer prostate tissues from healthy ones. It is clear from the figure that RRSFS/1NN algorithm performs better than simple SFS/1NN with high TPR rate.

#### **8. Conclusion**

In this chapter, a Round Robin SFS/1NN algorithm is proposed for the classification of prostate needle biopsies using multispectral imagery. To achieve this, a set of features was computed over a wide range of visible wavelength and the results have indicated a significant increase in the classification accuracy with Round Robin technique with high TPR. A key characteristic of the proposed Round Robin approach is that different features are used for each binary classifier from multispectral images, thus producing an overall increase in the classification accuracy. In contrast, in a multiclass problem, the classifier tries to find only those features that distinguish all classes at once. RR SFS/1NN has also demonstrated the effectiveness of some texture and structural features to make difference between different classes which can be helpful for the pathologist. Finally, the algorithm is generic and can be used for different datasets from other pattern recognition areas.

#### **9. References**


The results are comparable or better than those obtained in other recent studies (Taher & Bouridane, 2006); this further demonstrates that our new proposed Round Robin technique results in an improved ability to distinguish cancer prostate tissues from healthy ones. It is clear from the figure that RRSFS/1NN algorithm performs better than simple SFS/1NN

In this chapter, a Round Robin SFS/1NN algorithm is proposed for the classification of prostate needle biopsies using multispectral imagery. To achieve this, a set of features was computed over a wide range of visible wavelength and the results have indicated a significant increase in the classification accuracy with Round Robin technique with high TPR. A key characteristic of the proposed Round Robin approach is that different features are used for each binary classifier from multispectral images, thus producing an overall increase in the classification accuracy. In contrast, in a multiclass problem, the classifier tries to find only those features that distinguish all classes at once. RR SFS/1NN has also demonstrated the effectiveness of some texture and structural features to make difference between different classes which can be helpful for the pathologist. Finally, the algorithm is

generic and can be used for different datasets from other pattern recognition areas.

Breiman, L. (1996). Bagging predictors. *Machine Learning,* vol. 24 , 123-140.

sets. *Proc. AAAI Fall Symp Relevance*, pp.37–39.

*Report HPL-2003-4, HP Laboratories*.

*in Computer Science*, vol. 1451, Springer, Berlin, 611-619.

Barshack, I.; Kopolovic, J.; Malik, Z. & Rothmann, C. (1999)Spectral morphometric

Bartels, P. H et al. (1998). Nuclear chromatin texture in prostatic lesions: IPIN and adenocarcinoma. *Anal. Quant. Cytol. Histol*., vol. 20, no. 15,pp. 389–396. Bouatmane, S.; Nekhoul, B.; Bouridane, A. & Tanougast C. (2007). Classification of Prostatic Tissues using Feature Selection Methods. *IFMBE Proceedings*, vol 16, pp 843-846. Boucheron, L. ;Bi Z.; Harvey, N.; Manjunath, B. & Rimm, D. (2007). Utility of multispectral

Clark T. D.; Askin, F. B. & Bagnell, C. R (1987). Nuclear roundness factor: A quantitative

the effect of tumor stage on usefulness. *Prostate*. vol. 10, no. 3, pp. 199–206. Dash, M. Liu, H. (1997). Feature Selection for Classification. *Intelligent Data Analysis*, vol. 1,

Davies, S. & Russell, S. (1994). NP-completeness of searches for smallest possible feature

Duda, R. O. Hart, P. E. & Stork, D.G. (2001). *Pattern Classification*. Hoboken, NJ: Wiley-

Duin, R.P.W. & Tax, D.M.J. (1998). Classifier conditional posterior probabilities. *Lecture Notes* 

Fawcett, T. (2003). ROC graphs: Notes and practical considerations for researchers. *Tech* 

characterization of breast carcinoma cells. *Brit. J. Cancer*, vol. 79, no. 9–10, pp. 1613–

imaging for nuclear classification of routine clinical histopathology imagery. *BMC* 

approach to grading in prostate carcinoma, reliability of needle biopsy tissue, and

with high TPR rate.

**8. Conclusion** 

**9. References** 

1619.

*Cell Biology*, 8(Suppl. 1):S8.

no. 3, pp. 131- 156.

Interscience.


**Part 3** 

**Therapeutic Novelties** 

