**3.3.3 Selenium Flat Panel**

A selenium detector utilizes a thin layer (100-200 μm) of amorphous selenium for x-ray absorption. When an x-ray is absorbed by this material, it causes some electrons in the selenium to be liberated. The now "freed electron" and its corresponding "hole" from its departure create an electron-hole pair. This electron-hole pair creates the signal (Pisano et al., 2004). When electrodes are placed above and below the selenium and an electric field is applied, this causes the charges to move toward the electrodes. The signal is collected by one of the electrodes that are composed of a large matrix of dels. The dels act as capacitors to store the charge. At the corner of each del is a TFT switch. The readout of the charge is performed in the same manner as for the phosphor flat panel detector.

A detector of this kind is produced by Hologic (Danbury, CT). The Hologic detector dels are 70 μm with 14-bit digitization. Anrad (St Laurent Quebec, Canada) produces another selenium flat-panel system with 85 μm dels. To increase the geometric efficiency of this type of detector and to have a del of 50 μm, Fujifilm Medical has developed an amorphous selenium detector that has two separate layers of selenium as seen in Figure 11. In the Fuji system, the upper layer of selenium absorbs the x-rays and produces the electron-hole pairs. The charge is then stored in each del. The lower selenium layer will transfer the stored charge to a set of readout lines and then it will be transferred to an amplifier and digitized (Yaffe, in Bick & Diekmann, 2010). The information from one del will be used to create the information corresponding to a pixel of the image. This system has a bit depth of 14.

#### **3.3.4 X-ray (Photon) Quantum Counting**

The x-ray quantum counting detector is another example of a direct radiography mammography system. However, unlike the other detectors described above, it functions on the principle that each individual x-ray quantum is counted regardless of its energy. The unique design and concept of this detector allows each del to produce an electronic pulse every time an x-ray interacts with it (Pisano et al., 2004). The pulses are then counted and will create the signal for that pixel. Advantages of this type of detector are that no noise is associated with energy conversion and no analog-to-digital converter is needed.

Two quantum-counting systems have been developed. The detector in the Phillips Medical (Germany) [previously Sectra] system uses crystalline silicon in its multiple detectors. The detector and collimated x-ray beam move in synchrony across the breast. The x-rays are absorbed by the crystalline silicon. The electron-hole pairs are collected in an electric field and shaped into a pulse and counted (Aslund et al., 2007) as seen in Figure 12. The Phillips

A major advantage of this scanned-slot system is the result of the x-ray beam being collimated and only part of the breast being imaged at a time. Consequently, the transmitted x-rays are not lost (scatter-to-primary ratio is reduced), resulting in a grid no longer being needed. Therefore, the dose should be reduced. A limitation of this type of system is that it

A commercial unit like this was originally marketed by Fischer Imaging Inc (Denver, CO) as seen in Figure 8. It has dels of 54 μm with digitization performed at 12 bits. Of interest, over a small area of the detector, data could be read out at 27 μm to provide a high-resolution

A selenium detector utilizes a thin layer (100-200 μm) of amorphous selenium for x-ray absorption. When an x-ray is absorbed by this material, it causes some electrons in the selenium to be liberated. The now "freed electron" and its corresponding "hole" from its departure create an electron-hole pair. This electron-hole pair creates the signal (Pisano et al., 2004). When electrodes are placed above and below the selenium and an electric field is applied, this causes the charges to move toward the electrodes. The signal is collected by one of the electrodes that are composed of a large matrix of dels. The dels act as capacitors to store the charge. At the corner of each del is a TFT switch. The readout of the charge is

A detector of this kind is produced by Hologic (Danbury, CT). The Hologic detector dels are 70 μm with 14-bit digitization. Anrad (St Laurent Quebec, Canada) produces another selenium flat-panel system with 85 μm dels. To increase the geometric efficiency of this type of detector and to have a del of 50 μm, Fujifilm Medical has developed an amorphous selenium detector that has two separate layers of selenium as seen in Figure 11. In the Fuji system, the upper layer of selenium absorbs the x-rays and produces the electron-hole pairs. The charge is then stored in each del. The lower selenium layer will transfer the stored charge to a set of readout lines and then it will be transferred to an amplifier and digitized (Yaffe, in Bick & Diekmann, 2010). The information from one del will be used to create the

information corresponding to a pixel of the image. This system has a bit depth of 14.

associated with energy conversion and no analog-to-digital converter is needed.

The x-ray quantum counting detector is another example of a direct radiography mammography system. However, unlike the other detectors described above, it functions on the principle that each individual x-ray quantum is counted regardless of its energy. The unique design and concept of this detector allows each del to produce an electronic pulse every time an x-ray interacts with it (Pisano et al., 2004). The pulses are then counted and will create the signal for that pixel. Advantages of this type of detector are that no noise is

Two quantum-counting systems have been developed. The detector in the Phillips Medical (Germany) [previously Sectra] system uses crystalline silicon in its multiple detectors. The detector and collimated x-ray beam move in synchrony across the breast. The x-rays are absorbed by the crystalline silicon. The electron-hole pairs are collected in an electric field and shaped into a pulse and counted (Aslund et al., 2007) as seen in Figure 12. The Phillips

performed in the same manner as for the phosphor flat panel detector.

requires a longer image acquisition time.

**3.3.4 X-ray (Photon) Quantum Counting** 

**3.3.3 Selenium Flat Panel** 

mode.

Fig. 11. An illustration of the Fuji FFDM system that uses two layers of selenium. (Courtesy of Fujifilm Medical, Stamford, CT).

system has a del size of 50 μm and a bit depth of 16. In the system by XCounter (Stockholm, Sweden), it uses a set of multiple linear detectors and scans across the breast in synchrony with a collimated x-ray beam similar to the Sectra system. However, it uses a pressurized gas as the x-ray absorber. The pulses of ions generated by the gas form the signal (Thunberg, in Antonuk & Yaffe, 2002). The XCounter system has the same del size and bit depth as the Sectra. Neither system uses a grid.

#### **3.3.5 Photostimulable Phosphor (PSP) System**

The last detector to be discussed is the PSP system which is a computed radiography (CR) system. CR systems have been in use in general radiography for many years and are based on the principle of photostimulable luminescence. More recently they have been developed and used in mammography. The CR mammography systems utilize a phosphor screen. Energy from x-ray absorption causes electrons in the phosphor crystal to be liberated from the matrix and captured and stored in "traps" in the crystal lattice (Pisano et al., 2004), as seen in Figure 13. The number of traps filled is proportional to the amount of absorbed x-ray signal.

Digital Mammography 95

The CR image, which is analog, is then read out by placing the screen in a reading device. The reading device will scan it with a red laser beam in one dimension as it moves through the reader in the orthogonal direction. The red laser beam will free the electrons from the traps and cause them to return to their original resting state in the crystal lattice. As they return to their original state, the electrons will pass between energy levels created in the crystal with certain materials. The selected materials incorporated into the crystal typically emit blue light that is proportional to the x-ray energy absorbed by the phosphor (Yaffe, in Bick & Diekmann, 2010). The emitted blue light is measured with a light-collecting system composed of a photomultiplier tube, selected optical filter to eliminate the red light from interfering with the measurement, and a photomultiplier tube. Because the PSP is not composed of physical dels, spatial resolution of the system is the result of the size of the laser spot (del size) and the distance between sampling measurement (pitch) (Pisano et al., 2004). To decrease scan time while increasing light collection efficiency, SNR and sensitivity, some PSP vendors utilize a double-sided (read from the top and bottom surface of the PSP)

Fig. 14. CR mammography dual optical collection system scanned with a red laser beam.

A system of this type was developed by Fujifilm Medical Systems (Stamford, CT). Its del size is 50 μm and has a bit depth of 12. There are other PSP systems developed and used

Unlike the DR systems, the CR system uses removable cassettes that are placed into a bucky tray. This system can be a cost saver to some institutions since they may be able to convert their current SFM machine to CR. For work-flow, there may be little change in daily routine

reading device as seen in Figure 14.

(Courtesy of Fujifilm Medical, Stamford, CT)

throughout the world that have similar del and bit depths.

Fig. 12. X-ray (Photon) quantum counting detector. Schematic of a silicon based photon counting detector. (Courtesy of Phillips Medical, Germany)

Fig. 13. Schematic demonstrating energy from x-ray absorption liberates electrons from the phosphor crystal and are captured and stored in "traps" in the crystal.

Fig. 12. X-ray (Photon) quantum counting detector. Schematic of a silicon based photon

Fig. 13. Schematic demonstrating energy from x-ray absorption liberates electrons from the

phosphor crystal and are captured and stored in "traps" in the crystal.

counting detector. (Courtesy of Phillips Medical, Germany)

The CR image, which is analog, is then read out by placing the screen in a reading device. The reading device will scan it with a red laser beam in one dimension as it moves through the reader in the orthogonal direction. The red laser beam will free the electrons from the traps and cause them to return to their original resting state in the crystal lattice. As they return to their original state, the electrons will pass between energy levels created in the crystal with certain materials. The selected materials incorporated into the crystal typically emit blue light that is proportional to the x-ray energy absorbed by the phosphor (Yaffe, in Bick & Diekmann, 2010). The emitted blue light is measured with a light-collecting system composed of a photomultiplier tube, selected optical filter to eliminate the red light from interfering with the measurement, and a photomultiplier tube. Because the PSP is not composed of physical dels, spatial resolution of the system is the result of the size of the laser spot (del size) and the distance between sampling measurement (pitch) (Pisano et al., 2004). To decrease scan time while increasing light collection efficiency, SNR and sensitivity, some PSP vendors utilize a double-sided (read from the top and bottom surface of the PSP) reading device as seen in Figure 14.

Fig. 14. CR mammography dual optical collection system scanned with a red laser beam. (Courtesy of Fujifilm Medical, Stamford, CT)

A system of this type was developed by Fujifilm Medical Systems (Stamford, CT). Its del size is 50 μm and has a bit depth of 12. There are other PSP systems developed and used throughout the world that have similar del and bit depths.

Unlike the DR systems, the CR system uses removable cassettes that are placed into a bucky tray. This system can be a cost saver to some institutions since they may be able to convert their current SFM machine to CR. For work-flow, there may be little change in daily routine

Digital Mammography 97

Fig. 15. Digital mammogram a) before and b) after flat-filed correction. Courtesy of Martin

Mammography, eds. ED Pisano, MJ Yaffe, CM Kuzmiak. Lippincott, Williams & Wilkins, a

Fig. 16. Ghost or lag artifact. a) Artifact seen on a detector. b) Another example of ghosting. c) Clinical image with the appearance of a second breast (arrows). (Images **b** & **c** are

courtesy of Elizabeth Franklin of Carolinas Medical Center, Charlotte, North Carolina, USA)

Other digital artifacts can be attributed to the digital detector. A single non-functioning pixel may not be noticed or even affect clinical images, but if the numbers of non-functioning pixels are too many it will impact clinical images. The non-functioning pixel can appear as a white dot and may simulate calcifications. If a pixel discharges too early it may result in a

Yaffe, PhD; Sunnybrook Health Sciences Centre, Toronto, Canada. (from Digital

Walters Kluwer Company, 2004. With permission.)

for technologists. For some mammography technologists, CR systems have a similar "feel" as SFM. Cassettes with the PSP are still placed into the bucky tray, the patient is positioned and imaged the same, and then the plate is placed into a reading device without having to go into a darkroom. The mammographic image obtained can be printed to hard copy or displayed on softcopy just like the DR systems.

#### **3.4 Quality Control (QC)**

Mammography must be of high image quality in order for radiologists to detect the subtle changes of breast cancer. This can be difficult since breast cancer can present as a mass of the same density as normal breast parenchyma or be obscured by it. To ensure optimal performance from a system, the equipment must be properly set up and maintained. In the United States, the Mammography Quality Standards Act (MQSA) was established in 1992 to establish a federal mandated quality control program for screen film mammography. It now includes digital mammography. In Europe, as countries began to develop a breast cancer screening program, each began to develop their own national program. To standardize quality assurance and quality control, the European Commission published guidelines in 2006. Now in its 4th edition, the published guidelines include digital mammography (Perry et al., 2006). However, there is no single international source on QC procedures. In the United States, it is required that users follow the manufacturers' QC procedures for their systems.

Digital imaging allows the decoupling of acquisition, processing and display. In order to have meaningful QC of a digital system, each of the components must be evaluated separately. To evaluate acquisition (detector, beam, scatter, and radiation dose), quantitative measures are performed on the "raw" or unprocessed images. The QC for image processing is still in its early stages. To evaluate image display, electronic test patterns are used.

Image quality for mammography has been based on a test phantom, with objects imbedded within it. This test is subjective and prone to observer error. In the future, automated software may be useful for solving this problem (Young et al., 2008). Instead of a phantom, a more reliable measurement for verifying the consistency of image contrast in a digital system is the contrast-to-noise ratio (CNR). It is sensitive to changes in dose, object contrast and beam quality for each digital machine (Young, in Bick & Diekmann, 2010). The CNR object imaged is made of different materials that simulate the attenuation coefficient of breast tissue.

As with SFM, digital mammography can have artifacts. The artifacts may arise from detector non-uniformities. Over time, there can be degradation of the homogeneity of the detector leading to image degradation. To correct for this, an algorithm can be applied to all the images. The procedure is called "flat-field" or "gain correction" and is based on the principle that the detector responds linearly to radiation exposure. The first step in this test is to record the receptor response for the same amount of time of an image, but without any x-ray exposure. The values from this "dark" image are stored in the dels. In a subsequent image acquired using x-rays, the values stored are subtracted from the measurement from each corresponding del resulting in an image where it appears that the dark signals from all dels are zero (Yaffe, in Bick & Diekmann, 2010). Figure 15 is an example of flat-field correction.

for technologists. For some mammography technologists, CR systems have a similar "feel" as SFM. Cassettes with the PSP are still placed into the bucky tray, the patient is positioned and imaged the same, and then the plate is placed into a reading device without having to go into a darkroom. The mammographic image obtained can be printed to hard copy or

Mammography must be of high image quality in order for radiologists to detect the subtle changes of breast cancer. This can be difficult since breast cancer can present as a mass of the same density as normal breast parenchyma or be obscured by it. To ensure optimal performance from a system, the equipment must be properly set up and maintained. In the United States, the Mammography Quality Standards Act (MQSA) was established in 1992 to establish a federal mandated quality control program for screen film mammography. It now includes digital mammography. In Europe, as countries began to develop a breast cancer screening program, each began to develop their own national program. To standardize quality assurance and quality control, the European Commission published guidelines in 2006. Now in its 4th edition, the published guidelines include digital mammography (Perry et al., 2006). However, there is no single international source on QC procedures. In the United States, it is required that users follow the manufacturers' QC procedures for their

Digital imaging allows the decoupling of acquisition, processing and display. In order to have meaningful QC of a digital system, each of the components must be evaluated separately. To evaluate acquisition (detector, beam, scatter, and radiation dose), quantitative measures are performed on the "raw" or unprocessed images. The QC for image processing

Image quality for mammography has been based on a test phantom, with objects imbedded within it. This test is subjective and prone to observer error. In the future, automated software may be useful for solving this problem (Young et al., 2008). Instead of a phantom, a more reliable measurement for verifying the consistency of image contrast in a digital system is the contrast-to-noise ratio (CNR). It is sensitive to changes in dose, object contrast and beam quality for each digital machine (Young, in Bick & Diekmann, 2010). The CNR object imaged is made of different materials that simulate the attenuation coefficient of

As with SFM, digital mammography can have artifacts. The artifacts may arise from detector non-uniformities. Over time, there can be degradation of the homogeneity of the detector leading to image degradation. To correct for this, an algorithm can be applied to all the images. The procedure is called "flat-field" or "gain correction" and is based on the principle that the detector responds linearly to radiation exposure. The first step in this test is to record the receptor response for the same amount of time of an image, but without any x-ray exposure. The values from this "dark" image are stored in the dels. In a subsequent image acquired using x-rays, the values stored are subtracted from the measurement from each corresponding del resulting in an image where it appears that the dark signals from all dels are zero (Yaffe, in Bick & Diekmann, 2010). Figure 15 is an example of flat-field

is still in its early stages. To evaluate image display, electronic test patterns are used.

displayed on softcopy just like the DR systems.

**3.4 Quality Control (QC)** 

systems.

breast tissue.

correction.

Fig. 15. Digital mammogram a) before and b) after flat-filed correction. Courtesy of Martin Yaffe, PhD; Sunnybrook Health Sciences Centre, Toronto, Canada. (from Digital Mammography, eds. ED Pisano, MJ Yaffe, CM Kuzmiak. Lippincott, Williams & Wilkins, a Walters Kluwer Company, 2004. With permission.)

Fig. 16. Ghost or lag artifact. a) Artifact seen on a detector. b) Another example of ghosting. c) Clinical image with the appearance of a second breast (arrows). (Images **b** & **c** are courtesy of Elizabeth Franklin of Carolinas Medical Center, Charlotte, North Carolina, USA)

Other digital artifacts can be attributed to the digital detector. A single non-functioning pixel may not be noticed or even affect clinical images, but if the numbers of non-functioning pixels are too many it will impact clinical images. The non-functioning pixel can appear as a white dot and may simulate calcifications. If a pixel discharges too early it may result in a

Digital Mammography 99

Fig. 18. Processing artifact. Significant edge enhancement between the white breast tissue and black fat. This results in significant "blackness" of the image adjacent to the black white

interface (arrow).

bloom artifact, a white spot with a black halo secondary to the increase in charge of the neighboring dels (Van Ongeval, in Bick & Diekmann, 2010). If there is crystallization of the selenium of the detector, the images may become blurred over time (Marshall, 2006). Figure 16 are examples of ghost or lag artifact and are the result of incorrect electron clearing. Electronic interference can result in a zigzag artifact as seen in this CR image, Figure 17. Artifacts can be the result of image processing in which there is significant edge artifact between dense and fatty tissue, Figure 18. Also, artifacts can be related to the patient, i.e. chin, nose, finger, or hair artifact as seen in Figure 19.

Fig. 17. Electronic interference in a CR system resulting in a "zigzag" black line artifact.

bloom artifact, a white spot with a black halo secondary to the increase in charge of the neighboring dels (Van Ongeval, in Bick & Diekmann, 2010). If there is crystallization of the selenium of the detector, the images may become blurred over time (Marshall, 2006). Figure 16 are examples of ghost or lag artifact and are the result of incorrect electron clearing. Electronic interference can result in a zigzag artifact as seen in this CR image, Figure 17. Artifacts can be the result of image processing in which there is significant edge artifact between dense and fatty tissue, Figure 18. Also, artifacts can be related to the patient, i.e.

Fig. 17. Electronic interference in a CR system resulting in a "zigzag" black line artifact.

chin, nose, finger, or hair artifact as seen in Figure 19.

Fig. 18. Processing artifact. Significant edge enhancement between the white breast tissue and black fat. This results in significant "blackness" of the image adjacent to the black white interface (arrow).

Digital Mammography 101

In 1996 in the United States, the Federal Food and Drug Administration (FDA) published the *Information for Manufacturers Seeking Marketing Clearance of Digital Mammography Systems*  (Food and Drug Administration, 1996; Pisano, 2004). For manufacturers of digital mammography equipment seeking FDA-approval through 510(k) or Premarket Approval, the document required that each manufacturer demonstrate through a designed clinical trial that digital mammography equipment was equivalent to SFM. However, this was not without some challenges. The original FDA guidelines required manufacturers to demonstrate a higher rate of inter-reader agreement with FFDM than was obtainable between readers of SFM when SFM was compared to itself (Beam, 1996; Elmore, 1992; Howard, 1993). There was no requirement that manufacturers determine the truth about the presence or absence of breast cancer, only that the mammogram interpretations agree.

After an Advisory Panel met to discuss the "flawed" guidelines, the FDA released its revised guidelines on February 8, 1999. The guidelines now required approval trials to be based on breast cancer status truth. Consequently, sensitivity and specificity as measured by a Receiver Operating Characteristic (ROC) analysis was to be used with the goal of the studies to demonstrate that the difference in the areas under the ROC curve between digital and film was no greater than 0.1 (Pisano et al., 2004). This was now the FDA's standard for

Several minor clinical trials studies have been published in the US comparing digital mammography versus SFM, and these demonstrated promising results for digital mammography (Cole, 2001; Hendrick 2001; Pisano, 2004). One of the first major published clinical trials was the Colorado-Massachusetts Screening Trial (Lewin et al., 2001, 2002). The goal of this two-site study was to prospectively compare full-field digital mammography (FFDM) and SFM for cancer detection in a screening population. The design of the study was simple. All women at least 40 years old presenting for a screening mammogram were eligible to undergo both a SFM and a FFDM. A FFDM prototype system made by General Electric Medical (Milwaukee, WI) with a 18 cm x 23 cm amorphous silicon detector with a CsI crystal and a commercial SFM unit (DMR: General Electric Medical Systems, Milwaukee, WI) were used for imaging. A prototype softcopy display workstation was also

Final results by Lewin et al. were published in 2002 (Lewin et al., 2002). A total of 6,736 paired exams were performed on 4,521 women over a 30-month period. In these patients, 2,048 findings were detected in 1,467 of the studies with film, digital, or both imaging modalities. Additional work-up of the findings led to 183 biopsies and 42 were positive for cancer. The cancer detection rate was not statistically significant (p > 0.1). The difference between the ROC area for digital (0.74) and SFM (0.80) were not significant (p > 0.1). FFDM had fewer recalls than SFM (p < 0.001); however, the positive predictive values of both modalities were similar (3.3% SFM, 3.4% FFDM) (Lewin et al., 2002). Although FFDM did not lead to a higher cancer detection rate, it did lead to fewer recalls with a study that used a prototype unit and display workstation. This study becomes the foundation for the next

major screening clinical trial, Digital Mammography Imaging Screen Trial.

proving "substantial equivalence" of the two technologies.

**4. Clinical trials 4.1 United States** 

used.

Fig. 19. Hair artifact. White swirling lines representing the patient's hair are seen projecting over the medial portion of the breast.

Fig. 19. Hair artifact. White swirling lines representing the patient's hair are seen projecting

over the medial portion of the breast.

## **4. Clinical trials**

#### **4.1 United States**

In 1996 in the United States, the Federal Food and Drug Administration (FDA) published the *Information for Manufacturers Seeking Marketing Clearance of Digital Mammography Systems*  (Food and Drug Administration, 1996; Pisano, 2004). For manufacturers of digital mammography equipment seeking FDA-approval through 510(k) or Premarket Approval, the document required that each manufacturer demonstrate through a designed clinical trial that digital mammography equipment was equivalent to SFM. However, this was not without some challenges. The original FDA guidelines required manufacturers to demonstrate a higher rate of inter-reader agreement with FFDM than was obtainable between readers of SFM when SFM was compared to itself (Beam, 1996; Elmore, 1992; Howard, 1993). There was no requirement that manufacturers determine the truth about the presence or absence of breast cancer, only that the mammogram interpretations agree.

After an Advisory Panel met to discuss the "flawed" guidelines, the FDA released its revised guidelines on February 8, 1999. The guidelines now required approval trials to be based on breast cancer status truth. Consequently, sensitivity and specificity as measured by a Receiver Operating Characteristic (ROC) analysis was to be used with the goal of the studies to demonstrate that the difference in the areas under the ROC curve between digital and film was no greater than 0.1 (Pisano et al., 2004). This was now the FDA's standard for proving "substantial equivalence" of the two technologies.

Several minor clinical trials studies have been published in the US comparing digital mammography versus SFM, and these demonstrated promising results for digital mammography (Cole, 2001; Hendrick 2001; Pisano, 2004). One of the first major published clinical trials was the Colorado-Massachusetts Screening Trial (Lewin et al., 2001, 2002). The goal of this two-site study was to prospectively compare full-field digital mammography (FFDM) and SFM for cancer detection in a screening population. The design of the study was simple. All women at least 40 years old presenting for a screening mammogram were eligible to undergo both a SFM and a FFDM. A FFDM prototype system made by General Electric Medical (Milwaukee, WI) with a 18 cm x 23 cm amorphous silicon detector with a CsI crystal and a commercial SFM unit (DMR: General Electric Medical Systems, Milwaukee, WI) were used for imaging. A prototype softcopy display workstation was also used.

Final results by Lewin et al. were published in 2002 (Lewin et al., 2002). A total of 6,736 paired exams were performed on 4,521 women over a 30-month period. In these patients, 2,048 findings were detected in 1,467 of the studies with film, digital, or both imaging modalities. Additional work-up of the findings led to 183 biopsies and 42 were positive for cancer. The cancer detection rate was not statistically significant (p > 0.1). The difference between the ROC area for digital (0.74) and SFM (0.80) were not significant (p > 0.1). FFDM had fewer recalls than SFM (p < 0.001); however, the positive predictive values of both modalities were similar (3.3% SFM, 3.4% FFDM) (Lewin et al., 2002). Although FFDM did not lead to a higher cancer detection rate, it did lead to fewer recalls with a study that used a prototype unit and display workstation. This study becomes the foundation for the next major screening clinical trial, Digital Mammography Imaging Screen Trial.

Digital Mammography 103

The Oslo I study resulted in 3,683 women participating and a total of 31 cancers detected (detection rate 0.84%). Twenty-eight cancers were seen with SFM (detection rate 0.76%) and 23 by FFDM (detection rate 0.62%). The cancer detection rates were not significant (P= 0.23). The PPV for SFM was 46% and FFDM 39%. The recall rate after the consensus meeting was 3.5% (128 cases) for SFM and 4.6% (168 cases) for FFDM (Skaane et al., 2005). The recall rate was not significant. The authors concluded that there was no statistically significant difference in cancer detection rate; cancer conspicuity was equal between the modalities and soft-copy reading is comparable to SFM in a population based screening (Skaane et al.,

Other data to come out of the Oslo I study was a retrospective review of the cancers (Skaane et al., 2003). In this retrospective review, a side-by-side feature analysis of conspicuity of the cancers was performed and there was no difference between the modalities. In 2005, Skaane et al. published follow-up information on the missed FFDM cancers from the Oslo I study (Skaane et al., 2005). They concluded that inexperience of the readers with softcopy, improper viewing conditions and rapid interpretations might have contributed to the lower

Within a few months after completion of the Oslo I trial, patient enrollment for the Oslo II screening trial began. The aims of this study were to prospectively compare cancer detection rates, recall rates and positive predictive values of SFM versus FFDM in a screening program in Norway (Skaane & Skjennald, 2004). Women 50-69 years old were invited to the NBCS and women 45-49 years old were invited to the Oslo screening program. The patients were randomized for age and residence to undergo SFM or FFDM. Due to the physical location of the mammography equipment, the study investigators decided to have 70% of the patients undergo SFM and the other 30% to undergo FFDM (Skaane & Skjennald, 2004). The same radiologists who participated in the Oslo I trial participated. All images were again independently double-read (now with appropriate viewing conditions in the room) and scored using a 5-point malignancy scale. The potential "call back" cases were again

Results of the Oslo II study demonstrated a total of 64 cancers (cancer detection rate 0.38%) detected in 16,985 women who underwent SFM and 41 cancers (cancer detection rate 0.59%) in 6, 944 women who underwent FFDM (Skaane et al., 2007). The difference was in support of FFDM and was statistically significant (p = 0.03). The sensitivity was 77.4% for FFDM and 61.5% for SFM (P =0.07). The specificity was 96.5% and 97.9% respectively for the imaging modalities (P<0.005). The PPV was 15.1% for SFM and 13.9% for FFDM and this difference was not significant. The recall rate for SFM was 2.5% and 4.2% for FFDM (P < 0.001) (Skaane et al., 2007). The higher recall rate of this study confirmed the higher recall rate in the Oslo I trial. However, it is important to know that recall was based on a consensus conference, and it was only at the conference where comparison mammograms were made available for review. The study concluded that FFDM with softcopy interpretation is well suited for

Unlike the prior studies, the Swedish Helsingborg Study by Heddson et al. was a retrospective study that compared SFM, photon counting DR (PC-DR), and CR mammography from January 2000 to February 2005 (Heddson, et al., 2007). The goals were to evaluate cancer detection rates, recall rates, and PPV values in a screening population. A total of 52,172 screening mammograms were performed on 24,875 women during the study.

2005).

detection rate.

reviewed at a consensus conference.

breast cancer screening programs.

The Digital Mammography Imaging Screening Trial (DMIST) was sponsored by the American College of Radiology Imaging Network (ACRIN) and the National Cancer Institute (NCI) (Pisano et al., 2005). It was a cooperative venture by 33 sites in two counties led by Dr. Etta Pisano. A total of 49,528 asymptomatic women presenting for a screening mammogram underwent both a digital mammogram and SFM in a random order by the same technologist on the same day. A total of five digital systems were used: Senographe 2000D (General Electric), SenoScan (Fischer Medical), Selenium Full-Field Digital Mammography System (Hologic), Digital Mammography System (Hologic) and Computed Radiography System for Mammography (Fuji Medical). The images were read independently by radiologists, one radiologist for each exam. The radiologist scored each study on a 7-point malignancy scale for ROC analysis and a BIRADS (American College of Radiology, 2003) final impression to guide clinical work-up. The patient was recalled for additional diagnostic imaging whether the film, digital and/or both screening modalities demonstrated an abnormality. Standard work-up at each institution was performed that may have led to a biopsy. Breast cancer status was based on a breast biopsy within 15 months of study entry or a follow-up mammogram acquired at least 10 months after study entry.

Of the total patients recruited to DMIST over its 2-year accrual period, data was complete for 42,760 (86.3%) patients. A total of 335 breast cancers were diagnosed. Of these, 254 (75.8%) were diagnosed within 365 days and 81 (24.2%) were diagnosed between 366 and 455 days after study entry. The results demonstrate that for the entire population, the diagnostic accuracy of digital and film mammography was similar with a mean AUC of 0.78+/-0.02 for digital and 0.74+/-0.02 for film (difference in AUC, 0.03; 95 percent confidence interval, -0.02 to 0.08; P = 0.18). However, for certain subgroups, the accuracy of digital mammography was significantly higher than that of film mammography. The subgroups include women under the age of 50, women with heterogeneously dense or dense breasts, and premenopausal or perimenopausal women (P values of P =0.002, P = 0.003, P =0.002, respectively) (Pisano et al., 2005). Although the overall diagnostic accuracy of digital was similar to film, this large prospective, multicenter center study showed advantages for younger women with dense breasts.

#### **4.2 European clinical trials**

The first prospective digital screening study in Europe was performed in Norway in 2000 (Skaane et al., 2003, 2005). The Oslo I trial was a prospective study to compare SFM and FFDM with soft-copy reading in a population-based screening program. By invitation from the Norwegian Breast Cancer Screening Program (NBCSP), women aged 50-69 years of age were invited to participate. The women who agreed to participate in the study underwent two standard views of each breast with each modality (similar to the US studies). The FFDM studies were performed with Senographe 2000D (General Electric Healthcare, Buc, France) and interpreted on a GE softcopy display system with 2K x 2.5K monitors. Eight radiologists performed independent double readings for both modalities and used a 5-point malignancy scale. All images deemed positive (score of 2 or higher) were reviewed in a consensus meeting for each technique used. It was the consensus conference that decided whether the patient should be called back for additional imaging or scheduled for follow-up screening in 2 years.

The Digital Mammography Imaging Screening Trial (DMIST) was sponsored by the American College of Radiology Imaging Network (ACRIN) and the National Cancer Institute (NCI) (Pisano et al., 2005). It was a cooperative venture by 33 sites in two counties led by Dr. Etta Pisano. A total of 49,528 asymptomatic women presenting for a screening mammogram underwent both a digital mammogram and SFM in a random order by the same technologist on the same day. A total of five digital systems were used: Senographe 2000D (General Electric), SenoScan (Fischer Medical), Selenium Full-Field Digital Mammography System (Hologic), Digital Mammography System (Hologic) and Computed Radiography System for Mammography (Fuji Medical). The images were read independently by radiologists, one radiologist for each exam. The radiologist scored each study on a 7-point malignancy scale for ROC analysis and a BIRADS (American College of Radiology, 2003) final impression to guide clinical work-up. The patient was recalled for additional diagnostic imaging whether the film, digital and/or both screening modalities demonstrated an abnormality. Standard work-up at each institution was performed that may have led to a biopsy. Breast cancer status was based on a breast biopsy within 15 months of study entry or a follow-up mammogram acquired at least 10 months after study

Of the total patients recruited to DMIST over its 2-year accrual period, data was complete for 42,760 (86.3%) patients. A total of 335 breast cancers were diagnosed. Of these, 254 (75.8%) were diagnosed within 365 days and 81 (24.2%) were diagnosed between 366 and 455 days after study entry. The results demonstrate that for the entire population, the diagnostic accuracy of digital and film mammography was similar with a mean AUC of 0.78+/-0.02 for digital and 0.74+/-0.02 for film (difference in AUC, 0.03; 95 percent confidence interval, -0.02 to 0.08; P = 0.18). However, for certain subgroups, the accuracy of digital mammography was significantly higher than that of film mammography. The subgroups include women under the age of 50, women with heterogeneously dense or dense breasts, and premenopausal or perimenopausal women (P values of P =0.002, P = 0.003, P =0.002, respectively) (Pisano et al., 2005). Although the overall diagnostic accuracy of digital was similar to film, this large prospective, multicenter center study showed

The first prospective digital screening study in Europe was performed in Norway in 2000 (Skaane et al., 2003, 2005). The Oslo I trial was a prospective study to compare SFM and FFDM with soft-copy reading in a population-based screening program. By invitation from the Norwegian Breast Cancer Screening Program (NBCSP), women aged 50-69 years of age were invited to participate. The women who agreed to participate in the study underwent two standard views of each breast with each modality (similar to the US studies). The FFDM studies were performed with Senographe 2000D (General Electric Healthcare, Buc, France) and interpreted on a GE softcopy display system with 2K x 2.5K monitors. Eight radiologists performed independent double readings for both modalities and used a 5-point malignancy scale. All images deemed positive (score of 2 or higher) were reviewed in a consensus meeting for each technique used. It was the consensus conference that decided whether the patient should be called back for additional imaging or scheduled for follow-up screening in

advantages for younger women with dense breasts.

**4.2 European clinical trials** 

entry.

2 years.

The Oslo I study resulted in 3,683 women participating and a total of 31 cancers detected (detection rate 0.84%). Twenty-eight cancers were seen with SFM (detection rate 0.76%) and 23 by FFDM (detection rate 0.62%). The cancer detection rates were not significant (P= 0.23). The PPV for SFM was 46% and FFDM 39%. The recall rate after the consensus meeting was 3.5% (128 cases) for SFM and 4.6% (168 cases) for FFDM (Skaane et al., 2005). The recall rate was not significant. The authors concluded that there was no statistically significant difference in cancer detection rate; cancer conspicuity was equal between the modalities and soft-copy reading is comparable to SFM in a population based screening (Skaane et al., 2005).

Other data to come out of the Oslo I study was a retrospective review of the cancers (Skaane et al., 2003). In this retrospective review, a side-by-side feature analysis of conspicuity of the cancers was performed and there was no difference between the modalities. In 2005, Skaane et al. published follow-up information on the missed FFDM cancers from the Oslo I study (Skaane et al., 2005). They concluded that inexperience of the readers with softcopy, improper viewing conditions and rapid interpretations might have contributed to the lower detection rate.

Within a few months after completion of the Oslo I trial, patient enrollment for the Oslo II screening trial began. The aims of this study were to prospectively compare cancer detection rates, recall rates and positive predictive values of SFM versus FFDM in a screening program in Norway (Skaane & Skjennald, 2004). Women 50-69 years old were invited to the NBCS and women 45-49 years old were invited to the Oslo screening program. The patients were randomized for age and residence to undergo SFM or FFDM. Due to the physical location of the mammography equipment, the study investigators decided to have 70% of the patients undergo SFM and the other 30% to undergo FFDM (Skaane & Skjennald, 2004). The same radiologists who participated in the Oslo I trial participated. All images were again independently double-read (now with appropriate viewing conditions in the room) and scored using a 5-point malignancy scale. The potential "call back" cases were again reviewed at a consensus conference.

Results of the Oslo II study demonstrated a total of 64 cancers (cancer detection rate 0.38%) detected in 16,985 women who underwent SFM and 41 cancers (cancer detection rate 0.59%) in 6, 944 women who underwent FFDM (Skaane et al., 2007). The difference was in support of FFDM and was statistically significant (p = 0.03). The sensitivity was 77.4% for FFDM and 61.5% for SFM (P =0.07). The specificity was 96.5% and 97.9% respectively for the imaging modalities (P<0.005). The PPV was 15.1% for SFM and 13.9% for FFDM and this difference was not significant. The recall rate for SFM was 2.5% and 4.2% for FFDM (P < 0.001) (Skaane et al., 2007). The higher recall rate of this study confirmed the higher recall rate in the Oslo I trial. However, it is important to know that recall was based on a consensus conference, and it was only at the conference where comparison mammograms were made available for review. The study concluded that FFDM with softcopy interpretation is well suited for breast cancer screening programs.

Unlike the prior studies, the Swedish Helsingborg Study by Heddson et al. was a retrospective study that compared SFM, photon counting DR (PC-DR), and CR mammography from January 2000 to February 2005 (Heddson, et al., 2007). The goals were to evaluate cancer detection rates, recall rates, and PPV values in a screening population. A total of 52,172 screening mammograms were performed on 24,875 women during the study.

Digital Mammography 105

Three of the European studies have been discussed above to give examples of designs and results. In total, there have been seven published European population-based screening studies comparing SFM and FFDM (Del Turco, 2007; Heddson, 2007; Sala, 2009; Skaane, 2005, 2007; Vigeland, 2008; Vinnicombe, 2009). Table 1 compares the results. It is important to keep in mind when reviewing the data that each of the studies is of different design. Only the Oslo studies were prospective studies. The Oslo I was a paired-designed as were the US studies. Double reading was performed in the European studies followed by consensus or arbitration meetings for positive studies. In the US studies, each modality was interpreted using an independent reader blinded to the results of the other modality. The Colorado-Massachusetts Screening Trial by Lewin et al. used a consensus conference, DMIST did not. In DMIST, if a single reader, regardless of modality, noted an abnormality then the patient was recalled. The recall rates reports vary in the studies; four of the European studies show a higher recall rate with digital (Del Turco, 2007; Skaane, 2005, 2007; Vinnicombe, 2009). In five of the studies, the cancer detection rate was greater for digital (Del Turco, 2007; Heddson, 2007; Skaane, 2007; Vigeland, 2008; Vinnicombe, 2009). Two were statistically significant: Oslo II and Helsingborg Study (Skaane, 2007; Heddson, 2007). From these results, the utilization of digital screening mammography in the Western European

Image processing is the application of applying mathematical operations to the "raw" digital image with the aim to visualize subtle abnormalities so that they can be perceived by the radiologist as seen in Figure 20. Image processing should be robust and reproducible. The need for extra manipulation of the images should be none or very minimal for the radiologist. In addition, the radiologist should feel confident in reviewing the images. The unique property of digital is that each component of the system can be optimized. A few of

Intensity windowing algorithms are based on the principle that each acts on individual pixels in the image. A small portion of the full intensity range of the image is selected and then remapped to the full intensity range of the display (Pisano et al., 2000). With this process, it allows for the selection of specific intensity values of interest. Consequently, dense normal tissue and abnormal tissue values are exaggerated to accommodate for their

Manual Intensity Windowing (MIW) is one of the versions of intensity windowing. It is operator dependent. The radiologist manually windows and levels the images on a display system. Several examples are demonstrated in Figure 21. This can be quite variable depending on the radiologist's image reading preference and experience. Another version is histogram-based. In this version, it allows the system to select a window to allow the full range of contrast across the part of the histogram representing any fatty, mixed, or dense portion. The advantage of HIW is that it can adapt to individual breast types. The third version is Mixture-Model Intensity Windowing (MMIW). It provides region-specific window settings: background, compressed and non-compressed fat, dense tissue, and

Countries and the United States continues to increase.

small differences. This may result in increased lesion conspicuity.

muscle (Pisano et al., 2000). Examples of these are demonstrated in Figure 22.

the basic methods used will be discussed.

**5. Image processing** 

Fifty percent of the studies were performed with SFM, 19% with photon counting DR system (Sectra, Sweden), and 31% with a CR system (Fujifilm, Japan). The age range of the patients was 46-74 years of age. Forty percent of the SFM cases and 65% of the FFDM cases were double read by two radiologists. Recall of the patient was based by consensus of the radiologists.

Results of Helsingborg study demonstrated a statistically significant difference in the cancer detection rate for PC-DR versus SFM (P= 0.01) (Heddson et al., 2007). The cancer detection rates were 0.31% (81/25,901) for SFM, 0.49% (48/9841) for PC-DR, and 0.38% (63/16,430) for CR. In contrast to the Oslo studies, this study demonstrated a significantly higher recall rate for SFM than digital. The recall rate was for SFM, PC-DR, and CR were 1.4%, 1.0%, and 1.0%. As a result of the higher cancer detection rate and lower recall rate, the PPV for digital was higher than film [(P< 0.001): PC-DR = 47%, CR = 39% and film = 22%] (Heddson et al., 2007).

In addition to the above results, the Helsingborg Study demonstrated that digital mammography provided a dose reduction compared to film mammography as expected by its linear response to x-ray (Heddson et al., 2007). PC-DR provided a 75% dose reduction and CR a 16% dose reduction. The average glandular dose was 0.28 mGy, 0.92 mGy, 1.1 mGy, for PC-DR, CR, and SFM, respectively. The authors concluded that given the advantages of digital mammography, it is a valid alternative to screen film mammography.


CELBSS = Central East London Breast Screening Service Study.

Table 1. Published European studies comparing screen film mammography (SFM) and full field digital mammography (FFDM) for recall rate, cancer detection rate, and positive predictive value.

Fifty percent of the studies were performed with SFM, 19% with photon counting DR system (Sectra, Sweden), and 31% with a CR system (Fujifilm, Japan). The age range of the patients was 46-74 years of age. Forty percent of the SFM cases and 65% of the FFDM cases were double read by two radiologists. Recall of the patient was based by consensus of the

Results of Helsingborg study demonstrated a statistically significant difference in the cancer detection rate for PC-DR versus SFM (P= 0.01) (Heddson et al., 2007). The cancer detection rates were 0.31% (81/25,901) for SFM, 0.49% (48/9841) for PC-DR, and 0.38% (63/16,430) for CR. In contrast to the Oslo studies, this study demonstrated a significantly higher recall rate for SFM than digital. The recall rate was for SFM, PC-DR, and CR were 1.4%, 1.0%, and 1.0%. As a result of the higher cancer detection rate and lower recall rate, the PPV for digital was higher than film [(P< 0.001): PC-DR = 47%, CR = 39% and film = 22%] (Heddson et al.,

In addition to the above results, the Helsingborg Study demonstrated that digital mammography provided a dose reduction compared to film mammography as expected by its linear response to x-ray (Heddson et al., 2007). PC-DR provided a 75% dose reduction and CR a 16% dose reduction. The average glandular dose was 0.28 mGy, 0.92 mGy, 1.1 mGy, for PC-DR, CR, and SFM, respectively. The authors concluded that given the advantages of digital mammography, it is a valid alternative to screen film mammography.

> **Recall Rate (%) SFM**

**Recall Rate (%) FFDM**

paired 3,683 3,683 3.5 4.6 0.71 0.54 20.2 11.8

randomized 16,985 6,944 2.5 4.2 0.38 0.59 15.1 13.9

spective 25,901 9,841 1.4 1.0 0.31 0.49 21.8 47.1

spective 14,385 14,385 3.5 4.3 0.58 0.72 14.7 15.9

spective 324,763 18,239 4.2 4.1 0.65 0.77 15.1 18.5

spective 31,720 8,478 4.4 4.8 0.65 0.68 14.6 14.3

spective 12,958 6,074 5.5 4.2 0.42 0.41 7.5 9.7

Table 1. Published European studies comparing screen film mammography (SFM) and full field digital mammography (FFDM) for recall rate, cancer detection rate, and positive

**Cancer Detection Rate (%) SFM** 

**Cancer Detection Rate (%) FFDM** 

**PPV (%) SFM** 

**PPV (%) FFDM** 

radiologists.

2007).

**Study Study** 

**Oslo I** Prospective,

**Oslo II** Prospective,

**Helsingborg** Retro-

**Florence** Retro-

**Vestfold** Retro-

**CELBSS** Retro-

**Barcelona** Retro-

predictive value.

**Design** 

**SFM Exams** 

CELBSS = Central East London Breast Screening Service Study.

**FFDM Exams** Three of the European studies have been discussed above to give examples of designs and results. In total, there have been seven published European population-based screening studies comparing SFM and FFDM (Del Turco, 2007; Heddson, 2007; Sala, 2009; Skaane, 2005, 2007; Vigeland, 2008; Vinnicombe, 2009). Table 1 compares the results. It is important to keep in mind when reviewing the data that each of the studies is of different design. Only the Oslo studies were prospective studies. The Oslo I was a paired-designed as were the US studies. Double reading was performed in the European studies followed by consensus or arbitration meetings for positive studies. In the US studies, each modality was interpreted using an independent reader blinded to the results of the other modality. The Colorado-Massachusetts Screening Trial by Lewin et al. used a consensus conference, DMIST did not. In DMIST, if a single reader, regardless of modality, noted an abnormality then the patient was recalled. The recall rates reports vary in the studies; four of the European studies show a higher recall rate with digital (Del Turco, 2007; Skaane, 2005, 2007; Vinnicombe, 2009). In five of the studies, the cancer detection rate was greater for digital (Del Turco, 2007; Heddson, 2007; Skaane, 2007; Vigeland, 2008; Vinnicombe, 2009). Two were statistically significant: Oslo II and Helsingborg Study (Skaane, 2007; Heddson, 2007). From these results, the utilization of digital screening mammography in the Western European Countries and the United States continues to increase.

#### **5. Image processing**

Image processing is the application of applying mathematical operations to the "raw" digital image with the aim to visualize subtle abnormalities so that they can be perceived by the radiologist as seen in Figure 20. Image processing should be robust and reproducible. The need for extra manipulation of the images should be none or very minimal for the radiologist. In addition, the radiologist should feel confident in reviewing the images. The unique property of digital is that each component of the system can be optimized. A few of the basic methods used will be discussed.

Intensity windowing algorithms are based on the principle that each acts on individual pixels in the image. A small portion of the full intensity range of the image is selected and then remapped to the full intensity range of the display (Pisano et al., 2000). With this process, it allows for the selection of specific intensity values of interest. Consequently, dense normal tissue and abnormal tissue values are exaggerated to accommodate for their small differences. This may result in increased lesion conspicuity.

Manual Intensity Windowing (MIW) is one of the versions of intensity windowing. It is operator dependent. The radiologist manually windows and levels the images on a display system. Several examples are demonstrated in Figure 21. This can be quite variable depending on the radiologist's image reading preference and experience. Another version is histogram-based. In this version, it allows the system to select a window to allow the full range of contrast across the part of the histogram representing any fatty, mixed, or dense portion. The advantage of HIW is that it can adapt to individual breast types. The third version is Mixture-Model Intensity Windowing (MMIW). It provides region-specific window settings: background, compressed and non-compressed fat, dense tissue, and muscle (Pisano et al., 2000). Examples of these are demonstrated in Figure 22.

Digital Mammography 107

Fig. 22. Different post-processing algorithms (a) SFM of a cyst. (b-g) Photographic magnifications of a digital mammogram process with MIW (b), HIW (c), MMIW (d), CLAHE (e), unsharp masking (f), and peripheral equalization (g) (from *RadioGraphics*.

Adaptive Histogram Equalization (AHE) is a spatial enhancement method that changes pixel value based on spatial content. When applied to an image, there is enhancement of each pixel in relation to its local area. In doing so, all the gray values occur at an equal frequency in the image and consequently, the contrast of the background may be enhanced at the loss of contrast in the breast tissue (Karssemeijer, in Bick & Diekmann, 2010). In addition to the tissue contrast being increased in the image, so is noise. To limit noise, Contrast-Limited Adaptive Histogram Equalization (CLAHE) was developed. It limits the

Pisano et al. 2005. With permission).

Fig. 20. Digital mammogram. a) Raw image. b) Processed image. (Courtesy of Fujifilm Medical, Stamford, CT)

Fig. 21. Examples of different manual windowing and leveling of a digital mammogram that contains a spiculated mass. a) Initial. b) Manipulated image. c) Inverted image.

Fig. 20. Digital mammogram. a) Raw image. b) Processed image. (Courtesy of Fujifilm

Fig. 21. Examples of different manual windowing and leveling of a digital mammogram that

contains a spiculated mass. a) Initial. b) Manipulated image. c) Inverted image.

Medical, Stamford, CT)

Fig. 22. Different post-processing algorithms (a) SFM of a cyst. (b-g) Photographic magnifications of a digital mammogram process with MIW (b), HIW (c), MMIW (d), CLAHE (e), unsharp masking (f), and peripheral equalization (g) (from *RadioGraphics*. Pisano et al. 2005. With permission).

Adaptive Histogram Equalization (AHE) is a spatial enhancement method that changes pixel value based on spatial content. When applied to an image, there is enhancement of each pixel in relation to its local area. In doing so, all the gray values occur at an equal frequency in the image and consequently, the contrast of the background may be enhanced at the loss of contrast in the breast tissue (Karssemeijer, in Bick & Diekmann, 2010). In addition to the tissue contrast being increased in the image, so is noise. To limit noise, Contrast-Limited Adaptive Histogram Equalization (CLAHE) was developed. It limits the

Digital Mammography 109

radiologists read both conditions with at least one month between reads. Six cancers, 13 biopsy-proven benign lesions, 23 probably benign findings and 20 normal cases were included. The results demonstrated that softcopy display interpretations tended to be faster than film: mean time 34 seconds versus 40.5 seconds. In contrast, the ROC curve and sensitivity favored film (0.67:0.71 for film and 0.65:0.69 for softcopy). Specificity was slightly

With specificity of softcopy as a concern, a retrospective study comparing the specificity for calcifications in digital mammograms using softcopy versus film was performed by Kim et al. in 2006 (Kim et al., 2006). Eight radiologists reviewed 130 biopsy-proven cases of calcifications on softcopy and screen film. For each condition, the radiologists were asked to rate the probability of malignancy on a 5-point scale. For film, the radiologists could use a magnifying glass for further evaluation of the images, and for softcopy they could "roam and zoom" to manually window for contrast. The study concluded that there was no statistically significant difference in specificity achievable using softcopy digital versus

For radiologists using softcopy display for interpretations, appropriate room ergonomics and viewing conditions are absolutely necessary to minimize radiologist distractions and fatigue. Vendors need to continue developing hanging protocols and other tools that allow the radiologists to view all digital images with little manipulation of buttons or clicks of the

Digital mammography has transformed our everyday working environment. For many, gone are the days of darkrooms with wet chemical processing and the challenges associated with them. Film screen mammography view boxes are being used less each day as softcopy display becomes more familiar to the radiologist. Digital mammography images can be viewed from anywhere, and the trend for comparisons to be on softcopy is becoming apparent. Also, with prior studies being in digital format, the number of lost or missing priors has decreased. Files rooms with overcrowded shelves and stacks of folders piled on the floor are starting to disappear as picture archiving systems (PACS) with electronic

Digital mammography has also had a large impact on patient throughput. Technologist image acquisition and processing time with direct radiography (DR) digital mammography has been significantly decreased in both screening and diagnostic mammography (Berns, 2006; Kuzmiak, 2010). In the DR FFDM digital screening study by Berns et al., they studied the timed comparison of 183 hard-copy SFM cases and 181 FFDM softcopy display cases. Their results demonstrated a 7.5 min/case (35%) time savings over SFM (Berns et al., 2006). Results were similar in the diagnostic timed mammography study by Kuzmiak et al. (Kuzmiak et al., 2010). This prospective study consisted of 3 phases: 1st Phase, 100 patients imaged with SFM; 2nd Phase, 100 patients imaged on DR FFDM and interpreted on a recently installed softcopy display mammography system; 3rd Phase, same as 2nd Phase but 3-months after installation of the softcopy display system. Their results showed the

higher with softcopy (0.563 versus 0. 528), but not significant (Pisano et al., 2002).

screen-film mammography.

**7.1 Clinical workflow** 

storage are used.

**7. Digital imaging and clinical workflow** 

mouse.

maximum contrast adjustments by clipping and renormalizing the histogram (Pisano, 2000; Karssemeijer, 2010).

Unsharp Masking (UM) is a post-processing technique that is created by subtracting a lowpass filtered version of the original image from the original image (Chan et al., 1987). This process enhances high frequencies in the image such as calcifications and mass edges. A disadvantage of UM is that it also adds noise to the images. Also, it may falsely enhance a margin of a mass (i.e. an indistinct mass may appear more circumscribed and lead to inappropriate classification and follow-up of a mass instead of the need for a biopsy) (Pisano et al., 2000).

Peripheral Equalization or Peripheral Enhancement (PE) is a post-processing technique developed to improve the visualization of the less compressed (and over penetrated) outer edges of the breast. A filter is used to obtain a blurred version of the mammogram representing tissue thickness. This blurred image "mask" is scaled from 0 to 1, and the mammogram is divided by means of the mask values on a pixel-by-pixel bases (Byng et al., 1997). The applied algorithm acts on the pixels in the breast and where there is a thickness change. The result is that the pixels near the periphery are changed so the image becomes "flatter" across the mammogram and the periphery appears less black.

Image processing is a vital component of digital mammography. We may find that it may take more than one algorithm and not necessarily "film-like display" to evaluate for masses, distortion and calcifications. Image processing algorithms are currently not assessed as part of QA protocols. However, they should be to ensure the image chain is working optimally. Automated systems may be of great importance for efficient use of technologist and physicist time in QA and QC. There is a great need to continue to develop and evaluate this area of digital mammography.
