**4.2.2 How to calculate the blood glucose value from the spectrum**

We calculated the blood glucose value by using absorption spectra and PLSR. The invasion type blood glucose sensor (Antsense-2, DAIKIN) by enzyme electrode method is used for the measuring the reference blood glucose value.

First, we developed the calibration curve by PLSR, the measuring absorption spectra were used as the explanatory variable, measuring the blood glucose values by Antsense-2 were used as the objective variable. In calibration curve, the horizontal axis is the Antsense blood glucose value, and the vertical axis is the estimated blood glucose value by PLSR. Next, we measured to the new absorption spectra to measure the blood glucose value. Then, the blood glucose was calculated by using the measured absorption spectrum and calibration curve. And, we calculated the SEC and the SEP. Their values were used as the evaluated the accuracy of the calibration curve.

#### **4.2.3 Error Grid Analysis (EGA)**

EGA was developed by William L. Clarke in the University of Virginia. EGA is an indicator of clinical efficacy of blood glucose sensor. In fact, an error grid has been assembled to stay in the ideal range of 70 ~ 180 mg/dl blood glucose value.

The Clarke grid of EGA is shown in Fig.4. The horizontal axis is the actual blood glucose, and the vertical axis is the blood glucose value obtained from the developed blood glucose sensors. In this study, when measuring the blood glucose must be clinically effective. We judged the efficacy as a blood glucose sensor using the EGA.

#### **4.3 Absorption spectra of glucose, squalene oil, and finger**

In measuring the blood glucose value, must know the absorption peak of glucose spectrum.

In addition, we need to know the absorption wavelengths of squalene oil to be used as an internal standard method. Each sample was measured by ATR method (plate type prism). The absorption spectra of glucose powder and squalene oil are shown in Fig.5. The absorption spectrum of glucose has some absorption peaks in 1030 cm-1 (Fig.5-A, C-OH stretching vibration), 1130cm-1 (Fig.5-B, C-O-C antisymmetric stretching vibration), and 1450 cm-1 (Fig.5-C, CH2 scissoring vibration). The absorption spectrum of squalene oil has some

Fig. 4. Clarke grid

In measuring the subject's blood glucose value, the good accuracy calibration curve is essential. To improve the accuracy of the calibration curve must be accurately extract

In the ATR method, it is important to stick a sample to the prism. Many people of diabetes are elderly, person with dry skin on the fingertips are also often elderly. Squalene oil is used as internal standard method for the different dry skin of the subject's finger surface. To eliminate the effects of dry skin by applying the squalene oil, we can measure the subjects under the same conditions. And, there is no effect of squalene oil to apply normalization correction in the absorption peak of squalene oil. Furthermore, the S/N ratio of the spectrum is better, because to block the air from between the finger and prism by the oil.

We calculated the blood glucose value by using absorption spectra and PLSR. The invasion type blood glucose sensor (Antsense-2, DAIKIN) by enzyme electrode method is used for

First, we developed the calibration curve by PLSR, the measuring absorption spectra were used as the explanatory variable, measuring the blood glucose values by Antsense-2 were used as the objective variable. In calibration curve, the horizontal axis is the Antsense blood glucose value, and the vertical axis is the estimated blood glucose value by PLSR. Next, we measured to the new absorption spectra to measure the blood glucose value. Then, the blood glucose was calculated by using the measured absorption spectrum and calibration curve. And, we calculated the SEC and the SEP. Their values were used as the evaluated the

EGA was developed by William L. Clarke in the University of Virginia. EGA is an indicator of clinical efficacy of blood glucose sensor. In fact, an error grid has been assembled to stay

The Clarke grid of EGA is shown in Fig.4. The horizontal axis is the actual blood glucose, and the vertical axis is the blood glucose value obtained from the developed blood glucose sensors. In this study, when measuring the blood glucose must be clinically effective. We

In measuring the blood glucose value, must know the absorption peak of glucose spectrum. In addition, we need to know the absorption wavelengths of squalene oil to be used as an internal standard method. Each sample was measured by ATR method (plate type prism). The absorption spectra of glucose powder and squalene oil are shown in Fig.5. The absorption spectrum of glucose has some absorption peaks in 1030 cm-1 (Fig.5-A, C-OH stretching vibration), 1130cm-1 (Fig.5-B, C-O-C antisymmetric stretching vibration), and 1450 cm-1 (Fig.5-C, CH2 scissoring vibration). The absorption spectrum of squalene oil has some

**4.2.1 Internal standard method by squalene oil** 

the measuring the reference blood glucose value.

in the ideal range of 70 ~ 180 mg/dl blood glucose value.

judged the efficacy as a blood glucose sensor using the EGA.

**4.3 Absorption spectra of glucose, squalene oil, and finger** 

accuracy of the calibration curve.

**4.2.3 Error Grid Analysis (EGA)** 

glucose information on the infrared absorption spectrum.

**4.2.2 How to calculate the blood glucose value from the spectrum** 

Fig. 5. The absorption spectra of glucose powder and squalene oil

absorption peaks in 1377 cm-1 (Fig.5-D), 1462cm-1 (Fig.5-E), and 2922 cm-1 (Fig.5-F). The absorption peak of glucose and oil were not overlap.

The absorption spectra of the finger in before and after coating to squalene oil are shown in Fig.6. In absorption spectrum of finger, absorption peaks in 1377 cm-1 and 1462 cm-1 did not appear. But, absorption spectrum of finger coated with squalene oil has absorption peaks in 1377 cm-1 (Fig.6-D) and 1462 cm-1 (Fig.6-E). Therefore, it is possible to reduce individual differences in skin surface conditions due to the absorption spectra normalized at the absorption peak of squalene oil. We confirmed that the squalene oil is suitable as an internal standard method, and can measure the blood glucose value by absorption peaks of glucose.

#### **4.4 Non-invasive blood glucose measurement in subjects**

#### **4.4.1 Development of calibration curve**

The subjects were healthy eight men in their 20s. When developing a calibration curve, glucose tolerance test were performed on each subjects. In glucose tolerance test, first, the

Introduction of Non-Invasive Measurement Method by Infrared Application 87

E zone

(CC : Correlation coeficience, SEC : mg/dl)

All 0.79 21 77% 22% 0% 1% 0% #1 0.90 13 90% 7% 0% 3% 0% #2 0.95 16 90% 10% 0% 0% 0% #3 0.71 22 68% 32% 0% 0% 0% #4 0.71 14 83% 17% 0% 0% 0% #5 0.90 7 100% 0% 0% 0% 0% #6 0.84 16 90% 10% 0% 0% 0% #7 0.67 28 60% 35% 0% 5% 0% #8 0.96 9 95% 5% 0% 0% 0%

measurement accuracy of 6 among 8 subjects. In addition, the EGA results have improved in the same 6 subjects. The individual differences are eliminated by the individual calibration curve, and may be considered to be developed the most suitable calibration curve of each subject. Therefore, we measure blood glucose value of subjects by using the all subject

Each subject was measured again 20 times the infrared absorption spectra. We measured the blood glucose value by these absorption spectra and developing calibration curve in section 4.4.1. The result by calibration curve in all subjects and individual calibration curve are

In the result of blood glucose value calculated by all subjects calibration curve in Fig.8-(A), the SEP was ±32 mg/dl. This result was not good. On the other hand, in the result of blood glucose value calculated by individual calibration curve in Fig.8-(B) the SEP was ±13 mg/dl,

(A) The result by all subjects calibration curve (B) The result by #8 calibration curve

0

0 50 100 150 200 250

C

B

E B A

C

E

D

A

Reference value (mg/dl)

50

100

D

150

Estimated value (mg/dl)

200

250

Fig. 8. The result by developed calibration curve in all subjects and individual

E

D

A

0 50 100 150 200 250

C

B

E B A

C

Reference value (mg/dl)

Table 1. The results of all subjects and individual calibration curves by PLSR

**4.4.2 Non-invasive blood glucose measurement by calibration curve** 

shown in Fig.8. The average blood glucose value was 100 mg/dl.

calibration curve and individual calibration curve.

0

50

100

D

150

Estimated value (mg/dl)

200

250

CC SEC A zone B zone C zone D zone

Fig. 6. The absorption spectra of the finger in before and after coating to squalene oil

subjects fasted for 12 hours. Then, subjects are ingested 75 g glucose, temporary increases in blood glucose value of subject. After the glucose tolerance test, we measured to the infrared absorption spectra and the invasive blood glucose measurement in every three minutes 20 times. Furthermore, after breakfast, before and after lunch, before and after dinner, we measured for four days (five times measurement in a day). We measured the total 40 times in one subject. The calibration curve is developed using these infrared absorption spectra.

Fig.7 shows the calibration curve developed in 320 infrared absorption spectra by all 8 subjects. In these results, the average blood glucose of the subjects was 120 mg/dl, correlation coefficient (CC) was 0.79, and measurement accuracy (SEC) was ± 21 mg / dl. This result in this sample scale has been obtained a significant correlation, but, this result was not satisfactory as a blood glucose measurement system. As the cause of this, it is considered to the individual differences by moisture content of each subject. Therefore, we developed each individual calibration curve using the infrared absorption spectrum of each subjects. The results of individual calibration curves are shown in Table 1. Comparing the individual calibration curves and all subjects calibration curve has improved the

Fig. 7. The calibration curve developed by all subjects

E

Uncorting oil

Coating oil <sup>D</sup>

Fig. 6. The absorption spectra of the finger in before and after coating to squalene oil

4000 3500 3000 2500 2000 1500 1000 500

Wavenumber (cm-1)

0

1

Absorbance

2

Fig. 7. The calibration curve developed by all subjects

0

50

100

D

150

Estimated value (mg/dl)

200

250

subjects fasted for 12 hours. Then, subjects are ingested 75 g glucose, temporary increases in blood glucose value of subject. After the glucose tolerance test, we measured to the infrared absorption spectra and the invasive blood glucose measurement in every three minutes 20 times. Furthermore, after breakfast, before and after lunch, before and after dinner, we measured for four days (five times measurement in a day). We measured the total 40 times in one subject. The calibration curve is developed using these infrared absorption spectra. Fig.7 shows the calibration curve developed in 320 infrared absorption spectra by all 8 subjects. In these results, the average blood glucose of the subjects was 120 mg/dl, correlation coefficient (CC) was 0.79, and measurement accuracy (SEC) was ± 21 mg / dl. This result in this sample scale has been obtained a significant correlation, but, this result was not satisfactory as a blood glucose measurement system. As the cause of this, it is considered to the individual differences by moisture content of each subject. Therefore, we developed each individual calibration curve using the infrared absorption spectrum of each subjects. The results of individual calibration curves are shown in Table 1. Comparing the individual calibration curves and all subjects calibration curve has improved the

0 50 100 150 200 250

C

B

E B A

C

E

D

A

Reference value (mg/dl)


(CC : Correlation coeficience, SEC : mg/dl)

Table 1. The results of all subjects and individual calibration curves by PLSR

measurement accuracy of 6 among 8 subjects. In addition, the EGA results have improved in the same 6 subjects. The individual differences are eliminated by the individual calibration curve, and may be considered to be developed the most suitable calibration curve of each subject. Therefore, we measure blood glucose value of subjects by using the all subject calibration curve and individual calibration curve.

#### **4.4.2 Non-invasive blood glucose measurement by calibration curve**

Each subject was measured again 20 times the infrared absorption spectra. We measured the blood glucose value by these absorption spectra and developing calibration curve in section 4.4.1. The result by calibration curve in all subjects and individual calibration curve are shown in Fig.8. The average blood glucose value was 100 mg/dl.

In the result of blood glucose value calculated by all subjects calibration curve in Fig.8-(A), the SEP was ±32 mg/dl. This result was not good. On the other hand, in the result of blood glucose value calculated by individual calibration curve in Fig.8-(B) the SEP was ±13 mg/dl,

Fig. 8. The result by developed calibration curve in all subjects and individual

Introduction of Non-Invasive Measurement Method by Infrared Application 89

matched by changes in infrared absorption spectra of blood glucose variability. From the above, it has been shown to accurately measure the blood glucose value from the glucose information in this method. Therefore, we can be proposed as a non-invasive blood glucose

We have measured the diabetics that have been actually measured blood glucose. Subject is four diabetics. We were measured before and after meals for 4 to 5 days in each subject. First, we developed the individual calibration curve of each subjects by using the PLSR analysis and infrared absorption spectra obtained in the three-day measurement. Then, the measured infrared absorption spectra in the 1-2 day were substituted in the developed

Table 2 shows the blood glucose data for each subject obtained by the conventional method for three days from the first day, and the results of the developed calibration curve for each subject by using the measured infrared absorption spectra. And Fig.10 shows the calibration curve of subject #1. From these results, this measurement system can be constructed to calibration curve for measuring blood glucose of diabetics, because, significant correlation has been obtained in this sample scale. The result of subject #4 show very good results than the other subjects. As this cause, the number of samples is very few for construct calibration

Subject # Number Ave. Max. Min. S.D. C.C SEC 1 15 226 339 156 56 0.758 ± 35 2 16 216 330 150 44 0.519 ± 37 3 10 157 221 108 41 0.645 ± 29 4 9 113 203 58 44 0.997 ± 3

Table 2. The blood glucose data and the results of the calibration curve for each subject

C

0 50 100 150 200 250 300 350 400

<sup>A</sup> <sup>E</sup> <sup>B</sup>

E

D

B

A

Reference value (mg/dl)

C

measurement by the infrared spectral measurements.

**4.5 Clinical application of measurement system** 

Fig. 10. The calibration curve of subject #1

D

Estimated value (mg/dl)

calibration curve, and the blood glucose value was estimated.

this result was dramatically improved. And, by the EGA result, A-zone is 90%, and B-zone is 10%, this result was shown to be clinically effective. It is also considered to the impact of individual differences as this reason. Evanescent light is measured to invasion into the finger of about 0.3~2 μm in the ATR method, and, it is considered the infrared light not get to the blood vessel inside finger. Therefore, we are considered measured the interstitial tissue fluid. When the glucose comes out in tissue fluid from blood vessel, there is the time difference in between individual. So, if the subject measure the blood glucose by the calibration curve including other subjects information, the result have come to greatly affect these individual differences. For the above reasons, we measure the blood glucose value by using individual calibration curve of each subject.

#### **4.4.3 Verification of the calibration curve by loading vector**

In order to verify the validity of the calibration curve, we focused on the weighting of the loading vector obtained from the developed calibration curve. The loading vector is the calibration curve of data (infrared absorption spectra). In other words, the loading vector shows the wavenumber band to the impact to the calibration curve. The weighting of the loading vector by individual calibration curve in section 4.4.1 is shown in Fig.9-(A). The horizontal axis is the wavenumber, and the vertical axis is the weight of the loading vector.

There are absorption spectra three weighting curves because the number of PLS factors (principal components) obtained are three factors. In Fig.9-(A), in this loading vector, the wavenumber in 1220 cm-1 (Fig.9-(A)-A), 1150 cm-1 (Fig.9-(A)-B), 1020 cm-1 (Fig.9-(A)-C) were found to significant impact on the calibration curve.

Fig. 9. The weighting of the loading vector and each blood glucose value

Fig.9-(B) shows the infrared absorption spectra in large discrepancy between the subjects blood glucose value. In Fig.9-(B), characteristic absorption peaks can be found in 1220 cm-1 (Fig.9-(B)-A'), 1150 cm-1 (Fig.9-(B)-B'), 1020 cm-1 (Fig.9-(B)-C'). It is proportional to the blood glucose value and the absorbance in 1220 cm-1 and 1020 cm-1, and found a negative proportional relationship in 1150 cm-1. The weighting of the loading vector is almost matched by changes in infrared absorption spectra of blood glucose variability. From the above, it has been shown to accurately measure the blood glucose value from the glucose information in this method. Therefore, we can be proposed as a non-invasive blood glucose measurement by the infrared spectral measurements.

#### **4.5 Clinical application of measurement system**

88 Infrared Spectroscopy – Life and Biomedical Sciences

this result was dramatically improved. And, by the EGA result, A-zone is 90%, and B-zone is 10%, this result was shown to be clinically effective. It is also considered to the impact of individual differences as this reason. Evanescent light is measured to invasion into the finger of about 0.3~2 μm in the ATR method, and, it is considered the infrared light not get to the blood vessel inside finger. Therefore, we are considered measured the interstitial tissue fluid. When the glucose comes out in tissue fluid from blood vessel, there is the time difference in between individual. So, if the subject measure the blood glucose by the calibration curve including other subjects information, the result have come to greatly affect these individual differences. For the above reasons, we measure the blood glucose value by

In order to verify the validity of the calibration curve, we focused on the weighting of the loading vector obtained from the developed calibration curve. The loading vector is the calibration curve of data (infrared absorption spectra). In other words, the loading vector shows the wavenumber band to the impact to the calibration curve. The weighting of the loading vector by individual calibration curve in section 4.4.1 is shown in Fig.9-(A). The horizontal axis is the wavenumber, and the vertical axis is the weight of the loading

There are absorption spectra three weighting curves because the number of PLS factors (principal components) obtained are three factors. In Fig.9-(A), in this loading vector, the wavenumber in 1220 cm-1 (Fig.9-(A)-A), 1150 cm-1 (Fig.9-(A)-B), 1020 cm-1 (Fig.9-(A)-C) were

(A) The weighting of the loading vector (B) Absorption spectra Fig. 9. The weighting of the loading vector and each blood glucose value

C

Fig.9-(B) shows the infrared absorption spectra in large discrepancy between the subjects blood glucose value. In Fig.9-(B), characteristic absorption peaks can be found in 1220 cm-1 (Fig.9-(B)-A'), 1150 cm-1 (Fig.9-(B)-B'), 1020 cm-1 (Fig.9-(B)-C'). It is proportional to the blood glucose value and the absorbance in 1220 cm-1 and 1020 cm-1, and found a negative proportional relationship in 1150 cm-1. The weighting of the loading vector is almost


1300 1200 1100 1000 900

B'

75 mg/dl 134 mg/dl 171 mg/dl

C'

Wavenumber (cm-1)

0.0

0.2

Absorbance

0.4

0.6

A'

using individual calibration curve of each subject.

found to significant impact on the calibration curve.

A B

1750 1500 1250 1000 750

Wavenumber (cm-1)

vector.


0.0

Weighted factor

0.1

**4.4.3 Verification of the calibration curve by loading vector** 

We have measured the diabetics that have been actually measured blood glucose. Subject is four diabetics. We were measured before and after meals for 4 to 5 days in each subject. First, we developed the individual calibration curve of each subjects by using the PLSR analysis and infrared absorption spectra obtained in the three-day measurement. Then, the measured infrared absorption spectra in the 1-2 day were substituted in the developed calibration curve, and the blood glucose value was estimated.

Table 2 shows the blood glucose data for each subject obtained by the conventional method for three days from the first day, and the results of the developed calibration curve for each subject by using the measured infrared absorption spectra. And Fig.10 shows the calibration curve of subject #1. From these results, this measurement system can be constructed to calibration curve for measuring blood glucose of diabetics, because, significant correlation has been obtained in this sample scale. The result of subject #4 show very good results than the other subjects. As this cause, the number of samples is very few for construct calibration


Table 2. The blood glucose data and the results of the calibration curve for each subject

Fig. 10. The calibration curve of subject #1

Introduction of Non-Invasive Measurement Method by Infrared Application 91

number of the PLS factors. From the comparison of the SEP and SEC, the optimal number of PLS factors for this measurement system is considered to be 2 or 3. Thus, with less data in after the start of measurement, we shall construct the calibration curve careful in the number of PLS factors. To improve measurement accuracy, we need to consider improving the system. If increased in the amount of information obtained from absorption spectra, we can be measured blood glucose value more accurately. That is, if increased the amount of infrared light obtained

**4.6 Conclusion of non-invasive measurement for blood glucose by IR spectroscopy**  We can confirm non-invasive blood glucose measurement system developed in this study is effective tools in clinical practice. In this measurement system, there is no pain which was felt by many diabetics so far. Because measurement time is 1 minute, it can be measured very easily and quickly. There is no stress to the patient, can measure to blood glucose value

Future challenges of this study are shown below. The number of subjects and the number of measurement times must be increasing. We must show in this measurement reproducibility. And, we must show that this system can be applied to any subject. If the blood glucose measurement sensor has been developed in this measurement method, future, this research

**5. Non-destructive analysis of the composition and mixture ratio for textile** 

This section describes a Non-destructive analysis of the composition and mixture ratio for textile products. In Japan, the composition display of textile goods is obligated by the "Household Goods Labeling law". Therefore, analysis of the composition and mixture ratio for textile products in accordance with JIS method (L 1030-1,-2) are executed. But, existing methods is the destructive inspection, and a lot of time (six hours or more) and proficiency is necessary for the analysis. Also has concern about the safety and the environmental impact of such tests to use chemicals such as organic solvents. In addition, if the amount of remaining dirt on the sample is very small, if the remaining dirt color is same color of

In here, it proposes the application of the infrared spectroscopy for improve those problems. The method that we propose is non-destructive measurement, quickly and easy. We measured an infrared absorption spectrum of a textile by using a FT-IR. And we examined the possibility of a composition classification and a mixture ratio calculation by chemometrics. These experiments showed that the measurement system was effective.

Measuring system in this study, FT-IR (IRPrestige-21) and ATR method (DuraSampler-IR) was used. Measuring system is shown in Fig.12. The measuring yarn is folded, and set to cover on the prism. After, the measurement part of the prism and yarns were contacted by

**5.2 Analysis of the composition and mixture ratio for yarn products by ATR** 

from the subjects' finger, the measurement accuracy is improved.

several times a day because the measurement time is very short.

will contribute significantly in community of increasing diabetics.

**products based on infrared spectroscopy** 

sample, the dirt cannot be verified visually.

**5.2.1 Measurement system for yarn products** 

**5.1 Background** 

curves, and the average blood glucose value is small. In addition, it is considered to overfitting, because the PLS factor number is 5.

The spectra of diabetics as measured by the fourth day and fifth day were substituted in the individual calibration curve of each subject. Table 3 shows the accuracy of predicted blood glucose value and the EGA result. In addition, the scatter diagram of the Clark grid result of each subject is shown in Fig.11. From the EGA result, all data with one exception of subject #4 data are plotted in A and B zone of clinically safety ranges. Improvement of measurement accuracy is better, but, the important in clinical practice is give first aid when the blood glucose value showed abnormal. Therefore, from a good EGA results, this system can be treated as a reasonable measurement method of blood glucose value in diabetics in clinical practice.


Table 3. The accuracy of predicted blood glucose value and the EGA result

Fig. 11. The Clark grid result of each subject

To measure simply place a finger on the prism, this system has the advantage of measuring blood glucose value very easily for diabetics in many elderly. In addition, the burden on nurses is reduced, because the patient can measure in their own. Since the measurement time is about 1 minute, it is possible to measure the many patients in a short time. This is a worthwhile part in the clinical practice. When this system is used as a self-monitor blood glucose sensor of diabetic patients, the measurement error is a large. In particular, SEC of the subject #4 was ± 3 mg/dl, but SEP has become extremely large forecast error in ± 59 mg/dl. This cause is considered to the over-fitting, when building a calibration curve as previously described. In PLSR, if the increasing the number of the PLS factor, SEC is a smaller, SEP becomes large. Therefore, it is necessary to construct a standard curve with the optimal number of the PLS factors. From the comparison of the SEP and SEC, the optimal number of PLS factors for this measurement system is considered to be 2 or 3. Thus, with less data in after the start of measurement, we shall construct the calibration curve careful in the number of PLS factors. To improve measurement accuracy, we need to consider improving the system. If increased in the amount of information obtained from absorption spectra, we can be measured blood glucose value more accurately. That is, if increased the amount of infrared light obtained from the subjects' finger, the measurement accuracy is improved.

### **4.6 Conclusion of non-invasive measurement for blood glucose by IR spectroscopy**

We can confirm non-invasive blood glucose measurement system developed in this study is effective tools in clinical practice. In this measurement system, there is no pain which was felt by many diabetics so far. Because measurement time is 1 minute, it can be measured very easily and quickly. There is no stress to the patient, can measure to blood glucose value several times a day because the measurement time is very short.

Future challenges of this study are shown below. The number of subjects and the number of measurement times must be increasing. We must show in this measurement reproducibility. And, we must show that this system can be applied to any subject. If the blood glucose measurement sensor has been developed in this measurement method, future, this research will contribute significantly in community of increasing diabetics.
