International Conference on Communication and Signal Processing, July 28 - 30, 2020, India

Non-Invasive Glucose Monitoring using


Abstract—Diabetes Mellitus (DM), the metabolic disorder can cardio diseases [4]. To avoid these complications regular
cause serious health issues if not managed properly. The monitoring of blood glucose is very important, the current
conventional blood glucose monitors are invasive and causes pain
conventional method is invasive blood sampling [5], which
and discomfort to patients. Therefore, the study was aimed to
measure blood glucose non-invasively based on a machine can cause discomfort, pain after repeating use and even
learning technique. This system involves a Photoplethysmograph infection and tissue damage to patients [6]. This work use this
(PPG) based system using a light source of wavelength 525nm, fact of blood glucose concentration is proportional to the
660nm, and 950nm to determine the blood glucose parameters. transmitted light. There is research work that goes on using
The light source illuminates, the skin at the wrist and the biosensors with different methods like reverse iontophoresis
reflected beam is captured by a photodiode receiver. The detected
[7], Raman Spectroscopy [8], enzyme-based amperometric [9].
signal is conditioned, digitalized and passed to the Arduino UNO
microcontroller. The Arduino board derives the PPG signal in Thus the technology development tends to non-invasive but it
accordance with the subject blood glucose values. The raw has some problems to overcome and researches have reduced
waveform is pre-processed and subsequently segmented for and come with some novel ideas. They are sweat based sensor
obtaining the peak of the PPG signal. The random forest machine [10, 11], tear based contact lens [12], saliva-based [13]. There
learning technique is implemented on the acquired segmented are still some drawbacks in this method like calibration, long
signal, to obtain various statistical features namely mean,
settling time, isolation.
variance, skewness, entropy, kurtosis and standard deviation. The
machine learning system is designed and trained to estimate blood The rest of the paper is organized as follows. Section II
glucose from the extracted features. The blood glucose obtained describes the materials with its methodology. The system
from the proposed method, were comparable with the standard architecture is explained in Section III. The experimental
method. The results studied infers that the proposed method can procedures and results are discussed in Section IV and Section
be used for blood glucose monitoring. V respectively. At last, the paper concludes with conclusion of
the work.
Index Terms—Blood Glucose Level, Feature extraction,
Machine Learning, Near-Infra Red, Photoplethysmograph,
II. MATERIALS AND METHODOLOGIES
Signal acquisition, Signal analysis.
A. Working Principle
I. INTRODUCTION The skin is consists of 3 layers, they are epidermis, dermis,
hypodermis in which blood glucose information will be stored
D IABETES increase rapidly across the world and declared
a global epidemic by the World Health Organisation
in the dermis layer [14]. The near-infrared wavelength of
750nm to 2500nm can penetrate to dermis layer to obtain
(WHO) [1]. Diabetes has risen to 422 million in 2014 glucose information, so there is no need for invasive
according to WHO. Diabetes is a chronic disease, which is procedures [15]. The beam of NIR light is focused on the skin
caused by abnormal insulin levels in the body, due to not to absorb blood glucose information from the blood, the blood
insufficient amount of insulin produced by the pancreas or the absorbs more light than the tissues. Diabetic will have higher
cells not using it enough [2]. Insulin is a hormone that glucose molecules, so more amount of light is transmitted for
regulates glucose levels in the body. If the glucose level diabetic patients than non-diabetic patients [16]. Then the
remains very low for a long time can cause hypoglycemia and transmitted or reflected light is received using photodetector
hypoglycemia if it’s high [3]. Abnormal glucose levels for the [17]. The important thing in NIR is to use the right
long term can cause diseases like kidney problems, stokes and wavelength. Blood glucose known as glucose will be present
in S-H functional groups at the same time, other components
Hari Krishnan S is pursuing M.Tech Biomedical Engineering @ SRM
Institute of Science and Technology, Kattankulathur, Chennai – Tamilnadu,
like water and protein will be present in this hydrogen
India (e-mail: krishhari1996@gmail.com). functional groups. The most critical wavelength is the second
Vinupritha P is with the SRM Institute of Science and Technology, overtone band of 780-1400nm [18].
Kattankulathur, Chennai – Tamilnadu, India. (Phone: 91-8939680264, e-
mail: vinupritha@gmail.com).
Kathirvelu D is with the SRM Institute of Science and Technology,
Kattankulathur, Chennai – Tamilnadu, India. (e-mail: kathir297@gmail.com).

978-1-7281-4988-2/20/$31.00 ©2020 IEEE 0780

Authorized licensed use limited to: Cornell University Library. Downloaded on September 07,2020 at 07:05:43 UTC from IEEE Xplore. Restrictions apply.
 B. Photoplethysmograph Uno through the I2C interface. Arduino Uno will act as a
Photoplethysmography (PPG) is an optical technique used microcontroller unit (MCU) from which the dataset is sent into
to detect changes in volumetric changes of blood in the a personal computer for signal processing and glucose
microvascular bed of tissue. The change in blood volume by prediction.
pressure pulse is detected by transmitting NIR light into the
skin and then measured the amount of transmitted or reflected IV. EXPERIMENTAL PROCEDURES
light by the photodiode. The PPG uses a Beer-Lambert law for Use The output from the hardware circuit is recorded using
attenuation of light [19]. PLX-DAC software in a specific file as datasets. The ADC
values are stored and read in software platforms such as
(1) MATLAB, LABVIEW through excel files. The algorithm
through which the software code is implemented.
(2)
Fig. 2 shows that signal recorded is stored in different excel
files for each subject. These files are read to the software
Where, platforms for performing feature extraction and prediction
A = absorbance, ε = coefficient of absorption coefficient, model. As the signal is recorded for about three minutes, the
= optical path, C = the concentration of the light absorbing dominant region of the signal is segmented. The process of
component in the sample, = incident light intensity, = segmentation is done manually as the length of the datasets
transmitted light intensity, T = transmittance. may vary. A period of about one minute of the data is
A PPG waveform consists of a steady DC component and segmented and selected for further signal processing.
the pulsatile AC component. Constant DC light absorption is The segmented region of the signal is applied with software
due to pulsatile arterial and venous blood and AC light filters. Software filters such as wavelet transform, wavelet
absorption is due to the pulsatile volume of arterial blood [20]. decomposition, Savitzky-Golay filter are implemented in the
software. The process of removing unwanted frequencies in
III. SYSTEM ARCHITECTURE software using software filters is to additionally remove some
This system architecture explains that the transmitter used is noisy signals which might not be removed in hardware filters.
a green LED is 525nm, the red LED is 660nm, and NIR LED After removing noise frequencies from the signal, peak is
is 950 nm. The photodiode used is silicon photodiode of detected. Then feature extraction is done to obtain statistical
spectral range 0f 400nm – 1100nm. Filter used is a low pass features that are Mean, Variance Skewness, Entropy, kurtosis,
filter with a cut-off frequency of 0.3 Hz and a bandpass filter and standard deviation. After performing feature extraction,
with a cut-off frequency of 0.3Hz to 10Hz. The amplifier used using that statistical features classification tree is designed for
is a trans-impedance amplifier (TIA), which as low input estimation of glucose value using random forest machine
current and low voltage current. Arduino Uno is used as a learning technique.
microcontroller unit. PPG signal that is acquired in the wrist region. This signal
Fig. 1 shows the working methodology of non-invasive will be recorded, analysed and compared with the normal
blood glucose monitoring using machine learning. MIKROE subjects for estimating the prevalence of the stage of diabetes
2036 will have SFH 7050 and AFE 4404. SFH 7050 will in the patients. Further, it will also give an estimate of the
receive the reflected light in the form of current. AFE 4404 efficiency of the hardware design for the acquisition of the
will convert current it into voltage and transform it into a PPG signal. Besides, diagnosis is made by evaluating the
digital signal to Arduino sensitivity and specificity from the acquired data, thereby
comparing with the standard blood glucose testing methods.

Fig. 1. System Architecture Fig. 2. Proposed block diagram

0781

Authorized licensed use limited to: Cornell University Library. Downloaded on September 07,2020 at 07:05:43 UTC from IEEE Xplore. Restrictions apply.
 V. RESULTS Table I shows the blood glucose value obtained from
Fig. 3 is an input PPG signal recorded over a time period of proposed non-invasive method is compared with commercially
8 minutes using designed hardware. Fig. 4 is a segmented available invasive fully automated chemistry analyser. Results
signal of the input signal. An input signal of 8 minutes is shows from the comparison that proposed method as accuracy
segmented into a one minute signal using MATLAB software. of 94.23%. According to FDA and ISO, acceptable accuracy
Fig. 5 is a peak detection signal, which is obtained after range for glucose monitoring is ±15%. This results falls under
filtering the noise signal, the peak is detected for the signal this condition of clinically acceptable range for glucose
using MATLAB. monitoring. Thus, the proposed system is feasible for blood
glucose measurement non-invasively.

TABLE I
COMPARISON OF GLUCOSE MEASURES BETWEEN STANDARD AND
PROPOSED METHOD

Fig. 3. Input Signal

VI. CONCLUSION
In this research work, we have developed a hardware
prototype from which the Photoplethysmograph (PPG) signal
is acquired. The Integrated circuits used in this hardware
design is of low cost and has considerable advantages when
compared with the existing design used in the literature and
Fig. 4. Segmented Signal other existing devices. The signals that are acquired will be
analysed in software platforms for additional signal processing
methodologies and feature extraction will be performed for
both normal and diabetic subjects. The software analysis
highlights the feasibility of developing a low-cost non-invasive
glucose monitoring using machine learning. The future work of
this work will be focused on estimating the correlation of the
feature sets with various machine learning techniques.

REFERENCES
[1] N. Cho, J. Shaw, S. Karuranga, J. da Rocha Fernandes, A. Ohlrogge, B.
Malanda, IDF Diabetes Atlas: Global of diabetes prevalence for 2017
and projections for 2045, Diabetes Res, Clin. Pract. 138 (2018) 271,
https://doi.0rg/10.1016/j.diabres.2018.02.023.
[2] WHO, Diabetes, Available: (2020) http://www.who.int/news-room/fact-
Fig. 5. Peak Detection sheets/ detail/diabetes.

0782

Authorized licensed use limited to: Cornell University Library. Downloaded on September 07,2020 at 07:05:43 UTC from IEEE Xplore. Restrictions apply.
[3] Healthline: The Effects of Low Blood Sugar on Your Body. Available [12] H. Yao, Y. Liao, A.R. Lingley, A. Afanasiev, I. Lähdesmäki, B.P. Otis,
online: https://www.healthline.com/ health/low-blood-sugar-effects-on- B.A. Parviz, “A contact lens with integrated telecommunication circuit
body#6 (accessed on 2 October 2018). and sensors for wireless and continuous tear glucose monitoring”,
[4] A. Jina, M.J. Tierney, J.A. Tamada, S. McGill, S. Desai, B. Chua, A. Micromech. Microeng. 22 (2012) 075007,
Chang, M. Christiansen, “Design, development, and evaluation of a https://doi.org/10.1088/0960-1317/22/7/075007.
novel microneedle array based continuous glucose monitor”, J. Diabetes [13] J. Kim, S. Imani, W.R. De Araujo, J. Warchall, G. Valdés-ramírez,
Sci. Technol. 8 (2014) 483–487, T.R.L.C. Paixão, P.P. Mercier, J. Wang, “Wearable salivary uric acid
https://doi.org/10.1177/1932296814526191. mouthguard biosensor with integrated wireless electronics”, Biosens.
[5] Clark, L.C.; Lyons. C, “Electrode systems for continuous monitoring in Bioelectron. 74 (2015) 1061–1068,
cardiovascular surgery”, Ann. N. Y. Acad. Sci. 1962, 102, 29–45. https://doi.org/10.1016/j.bios.2015.07.039.
[CrossRef] [PubMed]. [14] K.A Unnikrishna Menon, Deepak Hemachandran, T K Abhishek “A
[6] So, C.-F.; Choi, K.-S.; Wong, T.K.S.; Chung, J.W.Y, “Recent advances survey on non-invasive blood glucose monitoring using
in noninvasive glucose monitoring. Med. Dev. (Auckl.) 2012, 5, 45–52. NIRSpectroscopy”, Communications and Signal Processing (ICCSP),
[CrossRef]. 2013 International Conference, April 2013, pp. 1069-1072.
[7] G. Rao, R.H. Guy, P. Glikfeld, W.R. LaCourse, L. Leung, J. Tamada, [15] Shailaja P Vedpathak, Dr.B.B.Godbole “Smart Ambulatory Blood
R.O. Potts, N. Azimi, “Reverse iontophoresis: noninvasive glucose Pressure Monitor: A Novel Approach”, International Research Journal
monitoring in vivo in humans”, Pharm. Res. 12 (1995) 1869–1873, of Engineering and Technology (IRJET), Nov 2015, vol 02, issue 08.
https://doi.org/10.1023/ A:1016271301814. [16] T. Tamura, Y. Maeda, M. Sekine and M. Yoshida, “Wearable
[8] R. Pandey, S.K. Paidi, T.A. Valdez, C. Zhang, N. Spegazzini, R.R. Photoplethysmographic Sensors-Past and Present”, Electronics, 2014
Dasari, I. Barman, “Noninvasive monitoring of blood glucose with vol.3, no.2, pp.282-302.
Raman spectroscopy”, Acc.Chem.Res. 50 (2017) 264–272, [17] R. W. Waynant “Overview of Non-Invasive Fluid Glucose Measurement
https://doi.org/10.1021/acs.accounts.6b00472. Using Optical Techniques to Maintain Glucose Control in Diabetes
[9] A.J. Bandodkar, W. Jia, C. Yardimci, X. Wang , J. Ramirez, J. Wang, Mellitus”, A Publication of The IEEE Lasers And Electro Optics
“Tattoo based noninvasive glucose monitoring : a proof – of - concept Society, 1998, vol.12.
study”, Anal. Chem. 87 (2015) 394–398, [18] Vega Pradana Rachim, Wan-Young Chung, Wearable-band type
https://doi.org/10.1021/ac504300n. visible-near infrared optical biosensor for non-invasive blood glucose
[10] W. Gao, S. Emaminejad, H.Y.Y. Nyein, S. Challa, K. Chen, A. Peck, monitoring, Sensors and Actuators: Chemical 286 (2019) 173-180.
H.M. Fahad, H. Ota, H. Shiraki, D. Kiriya, D.H. Lien, G.A. Brooks, [19] J.Liu, R.Liu, K.Xu, Accuracy of noninvasive glucose based near-
R.W. Davis, A. Javey, “Fully integrated wearable sensor arrays for infrared spectroscopy, Appl. Spectroscopy. 69 (2015) 1313–1318,
multiplexed in situ perspiration analysis”, Nature 529 (2016) 509–514, https://doi.org/10.1366/1407728.
https://doi.org/10.1038/nature16521. [20] Brince Paul, Melvin P Manuel, Zachariah C Alex, “Design and
[11] A. Abellán-Llobregat, I. Jeerapan, A. Bandodkar, L. Vidal, A. Canals, J. Development of Non-Invasive Glucose Measurement System”, IEEE
Wang, E. Morallón, “A stretchable and screen-printed electrochemical First International Symposium on Physics and Technology of Sensors,
sensor for glucose determination in human perspiration”, Biosens. 2012.
Bioelectron. 91 (2017) 885–891,
https://doi.org/10.1016/j.bios.2017.01.058.

0783

Authorized licensed use limited to: Cornell University Library. Downloaded on September 07,2020 at 07:05:43 UTC from IEEE Xplore. Restrictions apply.