Using an LED Light Source Coupled With Spectral Image Analysis for Non-Invasive Glucose Detection

Glucose monitoring is critical for diabetes patients. However, invasive blood testing carries the risk of infection if wound care is not handled properly and non-invasive glucose testing devices are often bulky and not portable. Therefore, our study proposes the use of white light-emitting diode (LED) bulbs with different color temperatures as a light source. Additionally, ammonium metavanadate and sulfuric acid were used to prepare the detection solution instead of peroxidase to produce color, after which images were captured using a smartphone. The red, green, and blue (RGB) channels were then separated and converted into grayscale images. The average grayscale value changes in the images were analyzed to determine the linear relationship between glucose concentration and grayscale values, thus allowing for non-invasive quantitative glucose testing. When using a 3000K white LED bulb as the light source, the grayscale values of the RGB channels analyzed from the images exhibited a linear relationship with the glucose concentration at a 0.1–10 mM range. The regression equation for the red channel was y = −3.1184x+148.2, with a coefficient of determination (R2) of 0.9165, a limit of detection (LOD) of 2.27 mM, and a limit of quantification (LOQ) of 6.88 mM. The proposed method of using white LED bulbs as a detection light source combined with image analysis can be used to determine whether glucose concentrations in blood, saliva, or tears are higher than normal, providing the advantages of rapid, highly sensitive, and non-invasive testing.

Using an LED Light Source Coupled With Spectral Image Analysis for Non-Invasive Glucose Detection Zhi Ting Ye , Senior Member, IEEE, Shen Fu Tseng , Shang Xuan Tsou , and Wen Tsung Ho Abstract-Glucose monitoring is critical for diabetes patients.However, invasive blood testing carries the risk of infection if wound care is not handled properly and non-invasive glucose testing devices are often bulky and not portable.Therefore, our study proposes the use of white light-emitting diode (LED) bulbs with different color temperatures as a light source.Additionally, ammonium metavanadate and sulfuric acid were used to prepare the detection solution instead of peroxidase to produce color, after which images were captured using a smartphone.The red, green, and blue (RGB) channels were then separated and converted into grayscale images.The average grayscale value changes in the images were analyzed to determine the linear relationship between glucose concentration and grayscale values, thus allowing for noninvasive quantitative glucose testing.When using a 3000K white LED bulb as the light source, the grayscale values of the RGB channels analyzed from the images exhibited a linear relationship with the glucose concentration at a 0.1-10 mM range.The regression equation for the red channel was y = −3.1184x+148.2, with a coefficient of determination (R 2 ) of 0.9165, a limit of detection (LOD) of 2.27 mM, and a limit of quantification (LOQ) of 6.88 mM.The proposed method of using white LED bulbs as a detection light source combined with image analysis can be used to determine whether glucose concentrations in blood, saliva, or tears are higher than normal, providing the advantages of rapid, highly sensitive, and non-invasive testing.

I. INTRODUCTION
T HE dynamic equilibrium of glucose in the body requires a rapid response to changes in plasma glucose concentration [1].Ideal plasma concentrations for medical purposes are usually within the range of 5.5-7.8mM, while blood glucose concentrations are typically within the range of 4.9-6.9mM [2].Previous studies have also demonstrated that excess glucose can damage cells and organ systems, whereas low glucose can impair neuronal function [3].Diabetes is a chronic disease that occurs when the pancreas cannot produce enough insulin or the body cannot effectively use the insulin it produces.This leads to hyperglycemia, which is the key characteristic feature of this disease [4].Once a diabetes diagnosis is established, patients are identified as having type 1 diabetes (T1D), where endogenous insulin production is nearly completely absent, or type 2 diabetes (T2D), where there is a defect in insulin sensitivity in energy-regulating organs [5].Studies conducted over the past three years have consistently demonstrated that diabetes has become a global health problem, with an expected increase in the number of patients in the future [6], [7], [8].The increase in diabetes cases can be attributed to several factors, including unhealthy eating habits, highly processed foods, lack of exercise, and obesity, among other problems caused by modernization [9], [10].High blood sugar levels can also lead to symptoms such as elevated eye pressure and even microvascular rupture, thus increasing the likelihood of retinopathy [11].Diabetes may also cause acute vascular blockage, increasing the risk of stroke [12], [13].The etiology of diabetes is multifactorial and the onset of this disease has been linked to many physiological stressors including genetics, lack of exercise, obesity, and poor eating habits [14], [15], [16].Therefore, early detection of high glucose levels is necessary to improve blood sugar control and reduce complications [17].Accurate measurement of blood sugar levels is crucial, as overconsumption of diabetes medication due to inaccurate blood sugar readings can lead to fatal consequences, including sudden death [18].
Handheld blood glucose meters with very high accuracy are widely available [19].However, not everyone can tolerate invasive treatment, especially pediatric patients.Therefore, there is an urgent need for non-invasive, painless alternative methods for monitoring blood glucose levels [20].According to medical research, the glucose concentration in the sweat of healthy individuals is approximately 0.277-1.11mM [21].The glucose concentration in saliva is approximately 0.238-0.716mM [22], and the glucose concentration in urine is approximately 2.78-5.55mM [23].
Spectroscopy is another non-invasive detection method for glucose detection [32], [33], [34], [35], [36], [37], [38].Spectroscopy consists of irradiating a light source onto a sample containing glucose, after which the amount of light passing through the sample is measured to determine the glucose concentration in the sample.However, current traditional detection equipment has better detection accuracy but typically requires a longer time for installation and operation.Additionally, their size and weight limit their mobility and maneuverability, and they may require specific facilities and infrastructure for use, making them unsuitable for real-time detection [39], [40].Miniaturized optical instruments for non-invasive glucose detection utilizing high-sensitivity flip-chip blue Mini-LEDs [41], this method cannot be determined through image analysis.
With the rapid advancement of technology, smartphones have become more than just communication tools and are now widely used as multi-functional devices.Recent studies have proposed the use of smartphones as detection tools for colorimetric and fluorescence measurements [42], [43], [44], [45], [46], [47], [48].However, the preparation of quantum dots often requires sophisticated technology and equipment, as well as relatively complex procedures.
Previous scholars have proposed a variety of DIB (Digital Image-Based) technology applications in biomedical testing [49].An approach utilizing fluorescence digital imaging with carbon quantum dots to assess the efficacy of a biocidal agent.[50].Proposing a novel method for detecting kidney dysfunctions: a smartphone-based digital image spot test for measuring uric acid levels in saliva [51].The number of articles utilizing the DIB method for chemical analysis has increased recently, owing to its ease of use, portability, rapid response, and extremely low energy consumption [52], [53].The number of articles utilizing the DIB method for chemical analysis has increased recently, owing to its ease of use, portability, rapid response, and extremely low energy consumption [54], [55], [56].
This study aimed to address the issue of inconvenient equipment for glucose detection by using white LED bulbs with different color temperatures (3000K, 4000K, and 6500K) as light sources, combined with image analysis methods for quantitative detection of glucose.Additionally, we proposed the use of ammonium metavanadate and sulfuric acid to prepare the detection solution instead of peroxidase to produce color.We also combined image analysis with smartphone camera imaging for detecting glucose color changes.Finally, a highly sensitive, fast, and detectable quantitative analysis method for glucose was achieved by analyzing the average grayscale value changes of images of different glucose concentrations.The results show that when using a 3000K white LED bulb as the light source and a smartphone camera as the detector, a coefficient of determination (R 2 ) of 0.9165 can be obtained within the concentration range of 0.1 to 10 mM.The detection limit (LOD) is 2.27 mM and the quantitation limit (LOQ) is 6.88 mM.
The method for detecting glucose was based on the results of two chemical reactions.The first step was to use glucose oxidase to catalyze glucose into hydrogen peroxide (H 2 O 2 ) and gluconic acid.Then, in the second step, ammonium metavanadate and sulfuric acid were separately added, causing a chemical reaction between hydrogen peroxide, ammonium metavanadate, and sulfuric acid, which finally led to the formation of a peroxovanadate complex The color image analysis proposed in this study consists of two parts.The first part involves capturing images of the cuvette after the completion of the colorimetric reaction.P(λ) and R(λ) represent the spectral distribution of the light source and the reflectance spectrum, respectively.Multiplying P(λ) and R(λ) yield the spectrum after the glucose reaction.The second part involves image analysis, where the image is divided into three channels, R, G, and B, using the MATLAB software and converted into a grayscale image for numerical analysis.The grayscale values of the RGB channels are divided by the same sampling pixel and the resulting average grayscale value is used to represent the color composition.Converting RGB images to grayscale images is achieved through the transformation of Y, U, and V, where Y represents luminance, U represents chrominance, and V represents chroma.
The formula for converting value of R, G, B to Y, U, V is given as (3): The detailed preparation method of the test reagent consisted of first adding 0.09 grams of glucose to 10 milliliters of water, Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.and stirring for 10 minutes using a magnetic stirrer at 500 rpm to completely dissolve the glucose.Then, the samples were diluted with water to a total volume of 50 milliliters, thus obtaining a glucose solution of the desired concentration of 10 mM.The glucose solution was then diluted with different amounts of water to prepare a total of 7 different concentrations of glucose solutions: 0.1, 0.4, 0.8, 1, 4, 8, and 10 mM.Next, a glucose oxidase solution with a concentration of 5 mg/mL was prepared.To simulate the body temperature during the process of glucose breakdown catalyzed by glucose oxidase, a beaker was heated over a hot plate with water to maintain the solution at 37°C.The temperature was then monitored using TES-1380 (TES Electrical Electronic Corp., Taipei, Taiwan) to ensure a constant temperature.After the catalysis was complete, a detection solution made by mixing 0.02 M ammonium metavanadate and sulfuric acid was added to complete the reaction.After the reaction, a peroxovanadate complex (NH 4 [VO(O 2 )SO 4 ]) is generated, and its normalized absorption and transmittance spectra are shown in Figure S1.
The analytical method used a coefficient of determination (R 2 ) to assess the linearity analysis of the color changes in the images and the concentration of glucose solution.This parameter is defined as the proportion of explained variance to total variance by the regression equation.The R 2 value ranges from 0 to 1, and the closer the R2 value is to 1, the more accurately the regression equation can explain the total variation.The calculation method for the coefficient of determination R2 is shown in (4): Where Ŷ is the predicted value of the regression model at point Y and Ȳ is the mean value of all Y values.
In this study, an iPhone 8 Plus was used for image capture.The distance between the camera lens and the cuvette (L 1 ) was fixed at 30 cm, and the white LED bulb was placed directly above the cuvette at a fixed distance of 110 cm (L 2 ).A black background was placed behind the cuvette, and the cuvette was placed on a horizontal experimental platform to ensure experimental stability and consistency.All photographs were taken in a darkroom throughout the experiment.A schematic diagram of the experimental setup is shown in Fig. 2(a).This study proposed using different color temperature white LED bulbs as a light source and image analysis with a smartphone camera to quantitatively detect glucose, as shown in the experimental flowchart in Fig. 2(b).First, glucose oxidase at a concentration of 5 mg/mL was allowed to react with seven different concentrations of glucose ranging from 0.1 mM to 10 mM at 37°C.Next, ammonium metavanadate and sulfuric acid were mixed with water to prepare the detection solution.The glucose, glucose oxidase, and detection solution were mixed at a 9:1:5 ratio.Afterward, all photographs were taken in a darkroom and three different color temperature white LED bulbs, 3000K, 4000K, and 6500K, were used as reference light sources in sequence.Finally, the grayscale values of different glucose concentrations were analyzed using the MATLAB software after taking the images with an iPhone 8 Plus smartphone and finally, the effect of different color temperatures on linearity results was analyzed.

III. RESULTS AND DISCUSSION
In this study, three white LED bulbs were used as light sources, and their normalized spectra.The black, red, and green lines represent the color temperatures of 3000K, 4000K, and 6500K white LED bulbs, respectively.
All images were acquired using an iPhone 8 Plus and the camera settings were adjusted to ensure that the photos were not overexposed.The shutter speed was set to 1/30 s, and the camera's ISO sensitivity was set to 400.After taking the photos, a fixed area of 150 × 150 pixels covering an area of 11 mm × 11 mm was captured from each cuvette, resulting in a total of 22500 pixels.Finally, the MATLAB software (R2019b) was used to convert the color pixels to grayscale and perform linearity analysis on the grayscale values.
First, a 3000K LED bulb was used as the light source, and the glucose solution samples captured with the iPhone 8 Plus camera in the glucose concentration range of 0.1-10.0mM are shown in Fig. 3. Fig. 3(a) shows the actual color image captured from the sample, which clearly shows that the color becomes darker as the glucose concentration increases.Fig. 3(b) shows the corresponding grayscale image obtained by converting the R channel of the image.It can also be seen that when the concentration is higher, the pattern color becomes darker.
The method of image processing with the MATLAB software (R2019b) involves dividing the captured image into three channels, R, G, and B, and conducting spectral analysis on the color image.Linear analysis was carried out using the average gray value and glucose concentration.The experimental results demonstrated that the linear relationship in the red channel was optimal when using a 3000K LED bulb as the light source and the iPhone 8 Plus camera as the detector, with a linear equation of y = −3.1184x+ 148.2 and an R 2 value of 0.9165, as shown in Fig. 3(c).The R 2 values for the G and B channels were 0.9094 and 0.8029, respectively.The LOD (3σ/S) was 2.27mM and the LOQ (10σ/S) was 6.88 mM.
Next, using a 4000K LED bulb as the light source, glucose solution samples were captured with the iPhone 8 Plus camera as shown in Fig. 4 for glucose concentrations ranging from 0.1 to 10.0 mM.Fig. 4(a) shows the color image captured from the actual sample, where it can be observed that as the glucose concentration increases, the color of the pattern becomes darker.Fig. 4(b) shows the grayscale image of the corresponding R channel when the glucose concentration was converted.Additionally, it can also be seen that the color of the pattern becomes darker as the concentration increases.
The experimental results showed that when using a 4000K white LED bulb as the light source and the iPhone 8 Plus camera as the detector, the linear relationship in the red channel was the best, with a linear equation of y = −3.1867x+ 166.42 and an R 2 value of 0.8977, as shown in Fig. 4(c).The R 2 values for the G and B channels were 0.8794 and 0.7827, respectively.
Finally, when using a 6500K LED bulb as the light source, the glucose samples captured with the iPhone 8 Plus camera in the concentration range of 0.1-10.0mM are shown in Fig. 4.   The experimental results show that when using a 6500K LED bulb as the light source and the iPhone 8 Plus camera as the detector, the linear relationship at the red channel is the best, with a linear equation of y = −2.5633x+ 161.97 and an R 2 of 0.9109.The linear relationship is shown in Fig. 5(c).The R 2 of the G and B channels are 0.8864 and 0.7869, respectively.
This study proposed the use of white LED bulbs with different color temperatures of 3000K, 4000K, and 6500K as light sources Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.

TABLE I COMPARISON OF COLOR IMAGE ANALYSIS RESULTS OF GLUCOSE WITH DIFFERENT DETECTION LIGHT SOURCES
to measure glucose solutions.The peak wavelengths were located at 600.5 nm, 446.9 nm, and 454.6 nm, respectively.We next analyzed the linear relationship between the average grayscale values of the RGB channels and glucose concentrations at a 0.1-10.0mM range, and the analysis results are shown in Table I.
The experimental results demonstrated that when using 3000K, 4000K, and 6500K LED bulbs as light sources and the iPhone 8 Plus camera as the detector, the linearity of the red channel at 3000K is the best among the three light sources, with an R 2 value of up to 0.9165.

IV. CONCLUSION
This study proposed using ammonium metavanadate and sulfuric acid to prepare the detection solution instead of peroxidase to produce color.Afterward, a smartphone was used for image capture and analysis for detecting changes in glucose concentration based on color shifts.
Next, we prepared a detection solution using ammonium metavanadate and sulfuric acid and used a white LED bulb as a light source combined with a smartphone camera for image capture.After capturing the images, they were processed and analyzed using the MATLAB software to quantify color differences by analyzing the average grayscale values of the RGB channels in the images to detect the glucose index.
The experimental results show that using a 3000K LED bulb as the light source and the iPhone 8 Plus camera as the detector resulted in better linearity.Within a glucose concentration range of 0.1-10 mM, the regression equation for the red channel was y = −3.1184x+148.2, with an R 2 value of 0.9165.The limit of detection (LOD) was 2.27 mM and the limit of quantification (LOQ) was 6.88 mM.
The proposed method offers advantages such as noninvasiveness, fast detection, and quantitative analysis.Moreover, it can detect concentrations in a range that includes blood, saliva, and tears, providing a more accurate analysis of glucose concentration changes and serving as a reference for diagnosis in medical institutions.

Fig. 1 .
Fig. 1.Schematic diagram of the image analysis process.

Fig. 2 .
Fig. 2. (a) Schematic diagram of the experimental architecture for quantitative detection of glucose with a white LED bulb with different color temperatures.(b) Experimental process of quantitative detection of glucose by image analysis of different color temperature white LED bulbs as detection light source.

Fig. 3 .
Fig. 3. (a) Color images of glucose with different concentrations.(b) Grayscale images of glucose with different concentrations in the R channel while using a 3000K LED bulb.(c) Linear relationship between different concentrations of glucose and average grayscale of the R channel with using a 3000K LED bulb.

Fig. 4 .
Fig. 4. (a) Color images of glucose with different concentrations.(b) Grayscale images of glucose with different concentrations in the R channel while using a 4000K LED bulb.(c) Linear relationship between different concentrations of glucose and average grayscale of the R channel with using a 4000K LED bulb.

Fig. 5 .
Fig. 5. (a) Color images of glucose with different concentrations.(b) Grayscale images of glucose with different concentrations in the R channel while using a 6500K LED bulb.(c) Linear relationship between different concentrations of glucose and average grayscale of the R channel with using a 6500K LED bulb.

Fig. 5 (
Fig. 5(a) is the color image of the actual sample, and Fig. 5(b) is the grayscale image of the R channel converted from the corresponding glucose concentration.It can also be seen that the color of the pattern becomes darker as the concentration increases.The experimental results show that when using a 6500K LED bulb as the light source and the iPhone 8 Plus camera as the detector, the linear relationship at the red channel is the best, with a linear equation of y = −2.5633x+ 161.97 and an R 2 of 0.9109.The linear relationship is shown in Fig.5(c).The R 2 of the G and B channels are 0.8864 and 0.7869, respectively.This study proposed the use of white LED bulbs with different color temperatures of 3000K, 4000K, and 6500K as light sources