2.1.

IBGDD Design

The major components of the IBGDD are the BGTS, a cover and baseplate set, a disposable lancet, and a light guide channel, as shown in Fig. 1. In our study, the smartphone’s LCD was used as the light source to capture the BGTS image. In order to guide the light to the BGTS area efficiently, after the IBGDD 3-D model was built using the computer-aided design software, an optical simulation using Tracepro (Lambda Research Corporation) was performed to find the optimal design for the IBGDD, especially the reflector’s angle.

Fig. 1

Smartphone and IBGDD, comprising the BGTS, cover, baseplate, light guide channel, and disposable lancet. The red arrows indicate the light path in the IBGDD.

2.2.

Simulation and Measurement Method for Achieving Uniform Illumination of the BGTS Area

The light source plays an important role in image analysis. To find an optimized RA that can guide light into the BGTS area, in this simulation experiment, the RA of the light guide channel was set from 30 deg to 70 deg with 5 deg intervals as the analysis conditions. In this manner, the most optimal angle for the best illuminance and illumination uniformity on the observation area of the BGTS could be determined. The actual illumination of the smartphone’s LCD was measured to be the light source illuminance value in the simulation. First, the screen was set to be white, and the screen brightness was raised to the maximum, and then a T10 illuminance meter (Konica Minolta, Japan) was used to measure the actual illumination of the screen. The average illumination is 230 lux. According to this measurement, the light source area was set to be $7.5\mathrm{mm}\times 6.3\mathrm{mm}$ in the simulation. The light wavelength was set by the general colorimetric method, in which the 550- nm wavelength is the most commonly used. The number of simulated traces was set to be 3 million.

The nine-point uniformity method was used to analyze the method of achieving uniform illumination in this study. The BGTS observation area was divided into nine symmetrical sampling areas, P1 to P9, as shown in Fig. 2. The sampling square size was $0.4\mathrm{mm}\times 0.4\mathrm{mm}$. The nine-area uniformity is defined by the minimum illuminance measured at the nine sampling areas divided by the maximum illuminance measured at the nine sampling squares, as

Fig. 2

Illuminance analysis of the BGTS area using the optical simulation program TracePro, with the BGTS area divided into nine symmetrical squares with sizes of $0.4\mathrm{mm}\times 0.4\mathrm{mm}$.

2.3.

Simulation Result

Our analysis of the simulation result indicates that the angle of reflection of the light guide channel has a significant effect on the illuminance and illumination uniformity of the BGTS observation area. At an angle of 30 deg, the illuminance and illumination uniformity are 50.43 lux and 90.43%, respectively. When the angle increases, both illuminance and illumination uniformity continuously increase. At 50 deg, the illuminance and illumination uniformity reach the highest values of 53.67 lux and 95.47%, respectively. With continued increase in the reflection angle, the illumination uniformity begins to decline. When the angle increases to 70 deg, the illumination uniformity declines to 90.68%, as shown in Fig. 3. It is concluded that when the RA of the light guide channel is too low, it cannot effectively reflect the light to the BGTS, and therefore both illuminance and illumination uniformity are low. However, when the angle is higher than 50 deg, the light intensity gradually decreases toward the BGTS observation area, and therefore the illumination and illuminance uniformity of the observed area will decline. According to the simulation, the final design of RA of light guide channel was selected to be 50 deg to achieve the optimal illuminance and illumination uniformity.

Fig. 3

Simulation results of illuminance and illuminance uniformity.

3.1.

Fabrication and Integration of the IBGDD

All IBGDD components were fabricated by plastic injection molding, because the process allows components with consistent characteristics at reasonable production cost. After fabrication, all components were cleaned by ultrasonic cleaning and then assembled together. The mold of the reflector channel in the reflective area was polished, which can ensure a mirror-like surface roughness and enhance the light reflection efficiency of the reflector.

In this experiment, generally marketed products of colorimetric blood glucose strip were selected as the BGTS material, which had concentrations ranging from 40 to 500 mg/dL. Since general strip dimension cannot meet IBGDD size a knife tool was used to cut the strip to the size $7.5\mathrm{mm}\times 6.3\mathrm{mm}$.

3.2.

Verification of the Illuminance and Illumination Uniformity of the BGTS Area

The illuminance measuring instrument used is Konica Minolta T10 illuminance meter. In order to verify the basic illuminance and illumination uniformity, a standard card of 4 mm diameter was used. The standard card has Pantone White and completely uniform color characteristics. After inserting the standard card into the IBGDD, the glucose measurement program was started and the smartphone’s LCD was switched on and guided to the BGTS area by the light guide channel, after which white balance locking and ISO value adjustment were performed. The program divided the BGTS area image into nine symmetrical sampling areas of square frame with four pixels each, and red, green, and blue (RGB) data from each rectangle image were collected to measure the minimum and maximum values. The measured data with the standard card were used to adjust the glucose measurement program for setting the basic illuminance and illumination uniformity. This operation was repeated five times to ensure the repeatability of the measurement.

3.3.

Preparation of Blood Glucose Samples with Different Concentrations

In this experiment, a Yellow Springs Instrument (YSI) biochemical blood glucose analyzer equipment (YSI-2300) was used to establish the different concentrations of blood glucose from the collected whole blood samples.

Sterile heparin tubes were used to collect 120 c.c. venous blood samples from each patient at environmental temperature of 22°C. After 24 h of standing, the blood glucose of the samples degraded to 0 mg/dL, and then proper amounts of glucose solutions were added to the samples without blood glucose according to the required blood glucose concentration. After complete mixing with a rotor equipment, the YSI-2300 blood glucose meter was used to measure the blood glucose concentration.

3.4.

Colorimetric Enzyme Reaction Analysis and Establishing the Signal Reference Mainline

The color of the colorimetric enzyme strip after reaction may vary with different blood glucose concentrations and reaction time. The first target of this experiment was to determine the signal of RGB specific and sensitive to the colorimetric enzyme strip and the time required to clearly distinguish the blood glucose concentration value of the samples. To avoid variations of blood glucose concentration caused by differences in temperature, the environmental temperature was set to 22°C. After inserting the strip into the IBGDD, the glucose measurement program was started and the image of the BGTS area was captured by the smartphone’s front camera (iPhone 5s, Apple Inc.) and the digital image of color change was separated into R, G, and B signals using the ColorAssist (FTLapps, Inc.) app according to the blood glucose concentration and reaction time. The concentration was adjusted from 50 to 500 mg/dL with an interval of 50 mg/dL. The experiment was repeated three times for each of the 10 different blood samples. Subsequently, the normalized algorithm was applied to obtain 10 different concentration curves of RGB signals, which could assist in determining the signal most suitable for the colorimetric strip, and the determined signal was used to establish linear equations with the parametric regression analysis method. Normalization was performed by dividing the reaction signal value by the unreacted signal. Given that the normalized signal values for each concentration were too low so as to easily distinguish them, the range of the normalized signal values for each concentration was rescaled up to 400 times. Finally, the linear equations were used as the signal reference mainline to convert the imaged strip colors to the blood glucose concentration values.

3.5.

Verification Plan for the Developed SMBG System

In 2013, ISO published a new standard ISO 15197:2013 and tightened the accuracy acceptance criteria.²³ The standard stipulates that to satisfy the minimum acceptable accuracy for a glucose monitoring system, 95% of measurement results must fall within $\pm 15\mathrm{mg}/\mathrm{dL}$ of the reference measured glucose concentration $<100\mathrm{mg}/\mathrm{dL}$ or within $\pm 15\%$ of the reference measured glucose concentration $\ge 100\mathrm{mg}/\mathrm{dL}$. To verify the accuracy of the new SMBG system within the criteria of ISO 15197:2013, a verification test was performed with venous blood from 20 diabetic patients. The diabetic patients in this verification plan covered diabetes types 1 and 2. The collected venous blood specimens were measured not only by the SMBG system but also by the YSI-2300 analyzer for accuracy comparison.

4.1.

Illuminance and Illuminance Uniformity of the BGTS Area

The T10 illuminance meter was used to measure the illuminance of the BGTS area and the average illuminance was 54.6 lux from five times of measurement. The nine-point uniformity method was used to measure and analyze the RGB signals with the standard card in the BGTS area. The measured uniformity values for RGB signals from five times of measurement were all higher than 94.5%. The values of G signal uniformity were significantly higher than those of the R and B signals, with an average value of 97.4%. The coefficient of variation (Cv) value of the RGB signals, ratio of the standard deviation to the mean value, was lower than 0.82%. The Cv of the G signal reached as low as 0.27%. This indicates that the designed reflector angle and reflection channel can provide uniform and stable light to the BGTS area. Uniformity and illuminance stability of the light source are extremely important factors affecting the accuracy and stability of the subsequent blood glucose measurement.

The measured illuminance and uniformity were also compared with the simulation results, and the comparison results proved the reliability of the simulation model. The simulation model will be used to improve the SMBG design in the future.

4.2.

Selection of the RGB Signal and Reference Mainline

The developed SMBG system was applied to record the RGB signal values every 0.2 s for 50 s, and each concentration sample was measured three times. The normalized algorithm was applied and 10 different concentration curves of RGB were obtained, as shown in Fig. 4. The figures show that the normalized curve of the R signal cannot be distinguished when the concentration is over 300 mg/dL [Fig. 4(a)], and the normalized curve of the B signal requires more than 30 s to distinguish each concentration curve [Fig. 4(b)]. In contrast, the 10 different concentration curves of the G signal could be distinguished completely after 10 s, as shown in Fig. 4(b). Therefore, the normalized curves of the G signal were selected to establish the signal reference mainline, and normalized curve values of each concentration at 10 s were collected. Subsequently, the parametric regression analysis method was used to obtain a linear equation $y=-0.3247x+219.25$, and the coefficient of determination $R^2$ was higher than 0.98, as shown in Fig. 5. The linear equation was used as the reference mainline to convert the imaged strip colors to the blood glucose concentration.

Fig. 4

Normalized curve from the measurement of 10 different concentration blood samples over 50 s. Each curve represents the average of three measurements. (a) The R signal cannot be distinguished when the concentration reaches over 300 mg/dL. (b) The 10 different concentration curves of the G signal can be clearly separated after 10 s. (c) More than 30 s is required to distinguish the concentration curves of the B signal.

Fig. 5

Signal reference mainline obtained based on the normalized value of the G signal with the parametric regression analysis method. Glucose concentrations of $50\sim 500\mathrm{mg}/\mathrm{dL}$ were recorded with the IBGDD and G signal and the results were read with a smartphone.

4.3.

Accuracy and Stability Tests for the Signal Reference Mainline

To verify the accuracy and stability of the signal reference mainline, six different blood samples with glucose concentrations adjusted at 50, 100, 200, 300, 400, and 500 mg/dL were tested and each sample was measured 10 times. The mean values of the samples from 10 times of measurement with the developed SMBG system were 49.0, 99.0, 198.3, 292.6, 397.0, and 488.7 md/dL, respectively; their standard deviations were 1.2, 2.9, 5.0, 9.6, 13.1, and 15.4, respectively; and their Cv values were 2.4%, 2.9%, 2.5%, 3.3%, 3.3%, and 3.1%, respectively. Compared with the adjusted glucose concentration of the samples, the accuracy of the measurements with the developed SMBG system was found to be 97.4%, 95.9%, 95.7%, 98.3%, 97.4%, and 95.8%, respectively. As the experimental results show accuracy values over 95% for all tested samples and all Cv values under 3.8%, the developed SMBG system with the established signal reference mainlines meet the standard criteria and clinical trial requirement for measuring blood glucose concentration with high accuracy and stability. The seven samples were also measured by the YSI-2300 analyzer for comparison. The concentration values were 50.4, 103.3, 207.3, 297.6, 407.8, and 510.2 mg/dL, respectively. Comparing these values, the developed system may measure the blood glucose concentration with the same level of accuracy and stability as the gold standard equipment.

4.4.

Verification Result for the Developed SMBG System

In the verification with venous blood collected from 20 diabetes patients, 4 were diabetes type 1 patients and 16 were diabetes type 2 patients. The age distribution is as follows: 5% under the age of 20, 15% between 21 and 30, 20% between 31 and 40, 25% between 51 and 60, 10% between 61 and 70, and 5% over the age of 71. According to the analysis of data from the 20 patients, measurement results of the developed SMBG system and YSI-2300 analyzer were compared, and the data were found to completely satisfy the $\pm 15\mathrm{mg}/\mathrm{dL}$ or $\pm 15\%$ criteria, as shown in Fig. 5. The accuracy acceptance percentage is 100% and meets the ISO 15197:2013 (E) criteria successfully. The parametric regression analysis of the data revealed a coefficient of determination ($R^2$) value of 0.9848, as shown in Fig. 6(b). These analyses show that the developed system measured the blood glucose concentration with the same level of accuracy and stability as the gold standard equipment. The design of the experiment and human subject involvement were approved by China Medical University Hospital.

Fig. 6

ISO 15197:2013 and parametric regression analysis results. (a) Comparison of the measured data from the developed SMBG system and the YSI-2300 analyzer for 20 samples. The plot also gives the superimposed tolerance bands according to the ISO 15197:2013 criteria for accuracy. (b) Parametric regression analysis plot comparing results for the SMBG and YSI-2300 analyzer.