Non-invasive glucose monitoring device

Abstract

The invention is a device providing non-invasive monitoring of glucose using array of visible / infrared sensors and machine learning approach.

Description

[0001] NON-INVASIVE GLUCOSE MONITORING DEVICE

[0002] Technical Field

[0003] The invention is a device providing non-invasive monitoring of glucose using array of visible / infrared sensors and machine learning approach.

[0004] State of the Art

[0005] Continuous monitoring of blood glucose is mandatory for diabetic patients. Today, the standard approach for measuring blood glucose is finger prick, but this method, which is applied by pricking the finger, is invasive and uncomfortable for patients. Also, the disposable glucometer sticks required for this approach are relatively costly. The finger prick method is based on the reaction of glucose oxidase with blood glucose producing gluconic acid and hydrogen peroxide (H2O2). The released hydrogen peroxide is detected by the electrochemical sensor and used to estimate the amount of glucose in the blood. Different studies to find a non-invasive approach for blood glucose level detection, including reverse iontophoresis, polarimetry, metabolic heat conformation, ultrasound, thermal emission, electromagnetic, photoacoustic, near-mid Infrared (IR) spectroscopy, Raman, light absorption and impedance spectroscopy is being carried out. It has been documented that with increasing blood glucose level, the mismatch between blood scatterings and the environment decreases, and therefore more light can pass through the blood sample. This is because the different speeds of light in different media cause the direction of the light to change. Frensel's law describes the change in light transmission for both transmittance and reflection. However, attempts to determine glucose level using near infrared light have been unsuccessful because many other parameters such as; skin color, hematocrit level, detector location, water, skin oil level etc. reduces the accuracy of glucose determination using near IR sensors.

[0006] US5222496 discloses non-invasive blood analysis method in which a blood sample is non- invasively illuminated through a patient's tissue, at a plurality of wavelengths, including a reference and a data wavelength, for to measure glucose concentrations. These wavelengths are preferably selected from the near infrared spectrum and are also preferably closely spaced to each other in order to minimize interference from other solutes in the blood. The used wavelength in this document is 1600 nm. US20170055891 discloses a device uses Near Infra-red (NIR) and mid infra-red (IR) thermal energy to determine blood glucose levels in a patient's body member where signals obtained only in reflectance mode.

[0007] Prior art documents do not mention using both transmittance and reflectance signals and also use of machine learning (multi-regression fitting).

[0008] Figures

[0009] Fig. 1: (a) Schematic explanation of the glucose level effect on light transmission. More glucose results in less scattering considering the Fresnel law and therefore more straight / shorter optical paths, and therefore less absorption. Please note that refractive index depends on light wavelength and the approximate refractive index values listed in figure are based on literature [4] for wavelength of 589 nm. (b) schematic figure showing the Fresnel law for transmission

[0010] Fig. 2: Schematic of the prepared device for collecting the training dataset and later for noninvasive monitoring of the glucose level of blood: design with 5 sensors and adjustable arm

[0011] Fig. 3: Schematic of electronic circuit of data acquisition setup

[0012] Fig. 4: Demonstration of computer program for prediction of glucose level using collected data and multiple linear regression approach (return glucose level button)

[0013] Fig. 5 : Step times of illuminations for data collection

[0014] Fig. 6: Flow Chart for data processing

[0015] M: Microcontroller

[0016] P: Phyton PC for further data processing, machine learning

[0017] SI: Color sensor

[0018] S2: Visible light detector

[0019] S3 : Infrared detector (850 nm)

[0020] S4: LDR sensor

[0021] S5: Infrared detector (940 nm)

[0022] 1: First portion

[0023] 2: Second portion

[0024] Detailed Description of the Invention

[0025] The device of the invention provides a non-invasive method for accurately determining glucose levels using a set of different light sources to determine the reflectance and transmittance of a body part with blood flow and a thickness below 2 cm, such as a finger or ear.

[0026] It has been well documented [1,2] that as the glucose level of blood increases, the mismatch between blood scatters and their surrounding media decreases, allowing more light to pass through the blood sample (Fig. la). This phenomenon is due to the different speed of light in different environments, which results in a change in the direction of light. The Fresnel law describes the change in light pass for both transmittance and reflectance, as shown in Fig. lb. More glucose results in less scattering considering the Fresnel law and therefore more straight / shorter optical paths, and therefore less absorption. Please note that refractive index depends on light wavelength and the approximate refractive index values listed in figure are based on literature [4] for wavelength of 589 nm. Fig lb shows the Fresnel law for transmission. Previous experiments of state of the art [3] have also supported this observation. However, the accurate and successful determination of glucose levels using near IR light has so far failed, as many other parameters, such as skin color, hematocrit level, position of detector, water, fat level of skin, and so on, can affect the transmittance of the finger, decreasing the accuracy of glucose determination using near IR sensors.

[0027] To overcome these limitations, the device of the invention comprises a set of different light sources at various wavelengths and sensors in different positions to collect a large database of glucose levels and corresponding sensor values. Once a database of sufficient size (size >1000 datapoint) is collected, advanced supervised machine learning algorithm is used, such as Multiple Linear Regression in Machine Learning, to train, validate, and test the model for glucose prediction. By utilizing machine learning, different light wavelengths and sensors in various positions, it is aimed to separate the interference of other parameters (such as skin color, hematocrit level, and other blood contents) from the glucose effect on transmittance since the transmittance / reflectance in different wavelengths differs for different parameters.

[0028] Multiple light sources are used in both transmittance and reflectance modes. The device comprises five sensors in different positions and with different sensitivity to different wavelengths of light. Transmittance and reflectance data measured by the sensors during five illumination steps for sheep ears (n=60) and rat legs (n=13) are collected, “n” is number of sheeps or rats. The real glucose level was also determined by a commercial finger-pricking (invasive) glucose monitoring device. The average data of the sensors were used to find the correlation between the measured sensor data and the real glucose level. After finding the five most sensitive sensor-illumination time combinations based on the goodness of fit of linear regression, multilinear regression analysis is performed to find a function that could convert measured sensor data to glucose level. An accuracy above 83% for rat hands and sheep with white ears is achieved, which is beyond the 80% state-of-the-art reported in the literature [1,2]. The device of the invention differs from prior art by using skin color sensor in reflection mode to increase the accuracy of glucose monitoring (LDR sensor in mirroring mode). Skin color has huge contribution in both transmittance and reflectance signal of used wavelength. Besides, visible light detector in transmittance mode (red light) is used. Red light in transmittance mode is used to have different data points and comparing this signal with others and using their interferences, to correlate the measured data with glucose level. Infrared detector with maximum sensitivity around 850 nm and 940 nm (coated LDR, photosensitive, connected resistive LDR) is also used. Additionally, the device of the invention measures the heartbeat signal by separate sensor and use it as additional signal to delete possible effect of blood flow rate on measured glucose level (ICQUANZX Heart Rate with built-in green LED monitor sensor).

[0029] Sensors used in the device:

[0030] • 5 different light sources from visible to near infrared region (450, 550, 650, 840 and 940 nm)

[0031] • SI color sensor for skin in reflectance mode

[0032] • S2 visible light detector in transmittance mode (red light)

[0033] • S3 infrared detector with maximum sensitivity around 850 nm (coated LDR- light dependent resistors LDR)

[0034] • S4 LDR sensor in reflectance mode (ICQUANZX Heart Rate Monitor sensor with embedded green LED)

[0035] • S5 Infrared detector with maximum sensitivity around 940 nm

[0036] Fig. 3 illustrated the electronic circuit of prepared instrument. Fig. 4 show the software developed for processing the data obtained by instrument.

[0037] Glucose level affects the transmittance and absorbance of used wavelengths, and it is used as an indicator to measure the glucose level. However, the transmittance and absorbance of used wavelengths are also affected by other parameters (e.g. skin color). The software of the invention reads and saves the measured reflectance and transmittance data by 5 sensors (Fig. 4). To be sure of maximum sensitivity of each sensor, 5 LEDs of instrument in different illumination intervals are turned on. As it is shown in Fig. 5, a total of 1000 iteration of measurement has been done and whole measurement was approximately 1 min. In the first interval which consists of the first 200 iterations, both blue and green light are on. In the next step, a red light turned on and this continue till all other LED turned on for 200 iterations. Note that since blue and green light mainly contribute to reflectance and are used to determine skin color, we turn them on at same time. Data from all five sensors are collected and saved during these steps for further processing.

[0038] The Device of invention comprises 2 portions. A first portion (1) and an opposing second portion (2) configured to receive a part of subject’s body therebetween (Fig. 2). First portion (1) comprises 5 different light sources from visible to near infrared region (450, 550, 650, 840 and 940 nm), color sensor for skin in reflectance mode (SI) and LDR sensor in reflectance mode (S4). Second portion (2) opposing the first portion (1) comprises visible light detector in transmittance mode (red light) (S2), infrared detector with maximum sensitivity around 850 nm (S3) and Infrared detector with maximum sensitivity around 940 nm (S5). First portion and second portion are used to receive part of the body therebetween. Part of the body can be finger or ear.

[0039] Data processing:

[0040] Prior to data processing an algorithm is used to eliminate noisy data obtained possibly due to movement during the measurements. The algorithm, check the standard deviation in middle of intervals (160 reading in middle checked) and if the standard deviation is bigger than 5% of average value, those measurement are rejected. The first and last 20 data of measurement in each iteration were not used to obtain average and standard deviation in each interval since LED and sensor response times can affect his part of collected data. In next step, to understand which data of sensor correlate with glucose level, first software plot average intensity of each sensor during each step of illumination (intervals) versus the glucose level.

[0041] Based on these plots, software determines sensitivity of each sensor for each light illumination step. Sensitivity data and collected goodness of fit (R2- goodness of fit indicator obtained from regression prediction) values used to select 5 most sensitive combination of sensors-illumination (Sx Ty: where x and y denote number of sensor and illumination intervals, respectively R2 closer to 1 used to select these 5 datasets). After this, multiple linear regression has been performed on data to correlate the 5 collected data with glucose level. In fact, having multiple sensors and light source enable the system to differentiate effect of other parameters of blood oxygen, heartbeat, and hematocrit level [2] as well as position of sensors in the setup. In final step, software read collected text file acquired by acquisition step and calculated the glucose level by using the coefficient of multiple linear regression.

[0042] Method for performing the device of the invention comprises steps of; i. Collecting sensor data of training set for 1 min by device of Claim 1 and saving the obtained data in a folder (large training set will results in high accuracy) ii. Collecting glucose level of training set iii. Checking the collected sensor data iv. If noisy, rejecting data and repeating measurement v. If successful data is collected then dividing data (80 % training, 10 % validation and 10 % testing data set) vi. Determining 5 most sensitive sensor-illumination time combination considering goodness of fit of linear regression vii. Analyzing multilinear regression using the determined combination and collected glucose level viii. checking the software using training, validation and testing data.

[0043] References

[0044] 1. Udara, S. SWI, et al. "DiabiTech-Non-Invasive Blood Glucose Monitoring System." 2019 International Conference on Advancements in Computing (ICAC). IEEE, 2019.

[0045] 2. Amir, Oma, et al. "Continuous noninvasive glucose monitoring technology based on “occlusion spectroscopy”." (2007): 463-469.

[0046] 3. Jafari, Ayyoob. "Artificial neural network assisted optical spectroscopy as a prospective tool for prediction of blood glucose level." Biointerface Research in Applied Chemistry (2016) Volume 6, Issue 5, 2016, 1559 - 1563

[0047] 4. Nanayakkara, Nuwan D., S. C. Munasingha, and G. P. Ruwanpathirana. "Non-invasive blood glucose monitoring using a hybrid technique." 2018 Moratuwa Engineering Research Conference (MERCon). IEEE, 2018.

[0048] 5. Richard H. ClarkeQian Wang, Infrared glucose sensor, US5222496A, 1992, United States

[0049] 6. Leila Chaychi, Method, device and system for non-invasive measurement of blood glucose content, 2015,US20170055891Al, United States.

[0050] 7. Hotmartua, Rolamjaya, et al. "Noninvasive blood glucose detection using near infrared sensor." 2015 International Conference on Electrical Engineering and Informatics (ICEEI). IEEE, 2015.

[0051] 8. Avner Gal Alexander M. Raykhman Eugene Naidis Yulia Mayzel Alexander Klionsky Anatoly Diber, Device for non-invasively measuring glucose, EP2563222B1, 2011, European Patent Office

Claims

CLAIMS1. A device for non-invasively measuring glucose level in a subject comprising.• 5 light sources of 450, 550, 650, 840 and 940 nm,• at least one-color sensor in reflectance mode (SI),• at least one visible light detector in transmittance mode (S2),• at least one infrared detector for 850 nm (S3),• at least one light dependent resistor in reflectance mode (S4)• at least one infrared detector for 940 nm (S5)2. Device according to Claim 1 characterized in that said infrared detector for 850 nm (S3) is coated light dependent resistor.

3. Device according to Claim 1 characterized in that said light dependent resistor in reflectance mode (S4) is Heart Rate Monitor sensor with embedded green light emitting diode.

4. Device according to Claim 1 comprises a first portion (1) and an opposing second portion (2) configured to receive a part of subject’s body therebetween.

5. Device according to Claim 5 characterized in that first portion (1) comprises 5 different light sources from visible to near infrared region (450, 550, 650, 840 and 940 nm), color sensor for skin in reflectance mode (SI) and LDR sensor in reflectance mode (S4) and second portion (2) opposing the first portion (1) comprises visible light detector in transmittance mode (red light) (S2), infrared detector with maximum sensitivity around 850 nm (S3) and Infrared detector with maximum sensitivity around 940 nm (S5).

6. Method of performing device of Claim 1 or 6 comprising steps of; i. Collecting sensor data of training set for 1 min by the device and saving the obtained data in a folder ii. Collecting glucose level of training set iii. Checking the collected sensor data iv. If noisy, rejecting data and repeating measurement v. If successful data is collected, then dividing data vi. Determining 5 most sensitive sensor-illumination time combination considering goodness of fit of linear regression vii. Analyzing multilinear regression using the determined combination and collected glucose level viii. checking the software using training, validation and testing data.

7. Method according to Claim 4 where successful data is collected then data is divided as 80 % training, 10 % validation and 10 % testing data set.

Images