10th International Conference on Electrical and Computer Engineering

20-22 December, 2018, Dhaka, Bangladesh 93

Non-invasive Blood Glucose Determination using

Near Infrared LED in Diffused Reflectance
Method
Mohammed Shahriar Arefin, 1,* Adnan Hossain Khan, 1 and Rabiul Islam1
1
Department of EEE, Chittagong University of Engineering and Technology
Raozan, Chittagong, Bangladesh
*
Shahriar.cuet2013@gmail.com

Abstract— Diabetes Mellitus is one of the notable health Glucoday etc. where a probe is inserted into the body, which
concerns around the globe. To keep diabetes in check a patient is also an uncomfortable method [10].
needs to track his/her blood glucose level regularly. In the To alleviate the patient’s discomfort, non-invasive
traditional invasive system, a patient needs to prick his body methods to determine the blood glucose level can be used.
part to collect the blood sample, which deems to be painful for
There are several methods of the non-invasive method
the patient. Non-invasive blood glucose measurement can
diminish the pain. In this paper, we proposed a non-invasive which include- Bio-impedance spectroscopy method, heat
blood glucose determination method where a NIR LED conduction method, Raman spectroscopy method, breathe
(940nm) and a photo-detector is used to determine the blood analysis method, mm-wave spectroscopy method, reverse
glucose concentration. We used the diffused reflectance iontophoresis method, ultrasound method, heart rate
method to determine the blood glucose concentration in our variability, mid-infrared spectroscopy, near-infrared (NIR)
paper. After implementing the device, we compared the data spectroscopy etc. [11]-[15], [6]. But there is no established
between our device and established invasive blood glucometer method that is being used globally due to dissatisfactory
available in the market. Clarke grid error analysis shows, that accuracy [16].
most of the measured data by the device are within Region-A.
Recently, many companies and researchers are trying
Keywords— Diffused Reflectance, Glucose level, NIR
near-infrared spectroscopy method to determine blood
LED, Non-invasive glucometer, Regression Analysis glucose level. The advantages of this method are- NIR
photo-detectors are low cost, widely available and have high
sensitivity. But the results in this method can be interfered
I. INTRODUCTION by variation in both physical parameters and environmental
The world is facing many major health problems and parameters. [17], [18]. In this NIR spectroscopy method,
diabetes or diabetes mellitus is one of them. Diabetes Robinson et al. first showed the relationship between blood
mellitus is a metabolic disorder where the blood glucose glucose level and NIR spectra, though the proposed
swings from normal blood glucose level (90-140 mg/dl) [1]. arrangement did not clarify the wavelength required for this
This chronic disease has high morbidity and it is established method [19]. Mauro et al. used optical fiber based NIR
that the disease is incurable [2], [3]. According to the system to correlate blood glucose level and NIR where they
International Diabetes Federation (IDF) survey report, indicated the potential of NIR spectroscopy as a method to
around 425 million adults were suffering from diabetes in determine blood glucose non-invasively [20]. Heinemann et
2017, they also assumed that the number will rise to 6 al. used optical fiber and a camera-based system where they
million in 2045 [4]. The survey shows there will be 48% measured the scattering coefficient. Their proposed system
more adults with diabetes in the next 28 years. The survey showed a significant relationship between near-infrared
also shows that 4 out 5 diabetes affected people live in low- scattering and blood glucose level [21]. Shyqyri et al. used
income countries [4]. NIR (940nm) transmittance spectroscopy to determine the
Type-1 diabetes mellitus has two variants; Hyperglycemia relationship between glucose level and transmitted light
and Hypoglycemia. Hyperglycemia is concerned with a high [22]. Yadav et al. used NIR (940nm) diffused reflectance
glucose level in the blood (>150 mg/dl) which can cause spectra in forearm where they showed the relationship
diabetic coma, blindness etc. [5]. On the contrast, between blood glucose and diffused reflectance [23].
Hypoglycemia is a condition in which the blood glucose Zhanxiao et al. proposed a multi-sensor based system where
level falls below a certain level (<60 mg/dl) [6]. Untreated they used NIR with the bio-impedance sensor and humidity
hypoglycemia can cause stroke, coma, confusion and sensor to determine blood glucose level by time series
irreversible brain damage [7]. To avoid complications, analysis [24]. All of the above-mentioned papers lack
blood glucose level needs to be checked on a regular basis. simplicity and are of high cost. Moreover, Most of the
Frequent checking of blood glucose can reduce the odds of devices are not suitable for wearable technology.
diabetic complications [8]. The purpose of this paper is to use near-infrared diffused
To monitor blood glucose level, there are two methods reflectance to implement a non-invasive blood glucose
which are being widely used nowadays. One is a finger stick measurement device. Our aim was to build the prototype at
glucose measurement system where a patient needs to prick the lowest possible cost along with simplicity. Our
his body part to take a blood sample. This method is painful, prototyped device uses 940nm NIR and also takes data from
can also induce fear among users as they need to prick their the finger which reduces the possible time delay problem
finger and repetitive use can cause infections [9]. Another that arises during the data collection. The device can be used
method is to use dynamic glucometer devices like- CGMS,

978-1-5386-7482-6/18/$31.00 ©2018 IEEE.

 94

as a wearable device for continuous blood glucose level of the capillary network [11]. Moreover, error due to time
measurement. delay while measuring the glucose level is minimized. The
data from photodetector then goes to Arduino for processing
II. THEORY and determination of blood glucose level. Fig. 2 shows the
According to beer-lamberts law, the light attenuation is overview of our proposed system.
proportional to the concentration of the medium. In the case
of human tissue, light interacts with tissue and attenuation
occurs [25]. The attenuation occurs due to absorption and
scattering of light. Fig. 1 depicts the scenario of attenuation
and transmittance.

Fig. 2: System Overview

B. Circuitry Design
At first, we designed a circuit as shown in Fig. 3.The the
LED and photo detector are kept at an angle of 45 degree in
Fig. 1: Attenuation of Light due to Scattering and
respect to finger and 90 degree from each other. The
Absorption [26]
distance between the LED and photo detector is 5mm. In
When glucose concentration increases, the absorption
this circuit, The NIR LED transmits light towards the finger.
increases but the scattering decreases. Due to a decrease in
scattering the diffuse reflectance also gets decreased. In a
semi-infinite medium, diffusion reflectance R(r) is
expressed as

(1)
Here,
Distance from isotropic source to detector=r,
Diffusion coefficient,
Absorption coefficient = µa
Scattering coefficient = µs
As, with concentration increase µa increases, thus δ
Fig. 3: Preliminary Circuit for determining Relationship
increases too. As the diffusion coefficient increases
between blood glucose level and light
diffusion reflectance gets reduced [27]. By applying this
The light gets reflected from the finger after interaction
relationship between diffusion reflectance and blood glucose
the blood glucose. The photo-detector collects the reflected
concentration, near infrared LED-based system can be
light. Finger placement displacement from the LED has
designed to determine the blood glucose level.
been set 2-3mm. The voltage across the voltage divider
III. SYSTEM STRUCTURE changes with variation of light. The voltage level is detected
by the Arduino and then convert data into ADC value. Later
A. System Design to reduce the possible ac line interference (50Hz) we
NIR region in the electromagnetic spectrum (800- introduced a 50Hz notch filter with the circuit. Fig. 4(a)
2500nm) includes three bands [28]. We are using second or shows the schematic of final circuitry and Fig. 4(b) depicts
higher overtone band (800-1200nm). The second overtone the picture of circuitry along the display LCD.
band is mainly concerned with scattering and due to shorter
wavelengths, the penetration depth in the skin decreases
[29], [30]. As we are using diffuse reflectance method so it
is convenient to select this region. We used market
available near-infrared LED of 940nm. Also, the
wavelength 940nm does not interfere with the absorption
peaks of water (1450nm, 1787 nm), protein (2174nm, 2288
nm) and fat (2299nm, 2342 nm) [31]. To detect the reflected
light after interaction with the human body. In our system,
we are taking data from the finger. The reason for taking
data from the finger is that it is hair free, has a high density (a)
 95

(b)
Fig. 4 -Final Circuitry: (a) Schematic Diagram and (b) Fig. 5: ADC Value vs. Glucose Level
Implemented Circuit
IV. TESTING AND ANALYSIS
We conducted our experiment on 10 people which
include both diabetic and non-diabetic person. We collected
glucose level data by NIR sensor and a commercial invasive
glucometer (Gluco Dr. Auto) at the same time. We took 55
data from patients during this project. The uniqueness of our
data collection is that we collected data at various times of
the day rather than the conventional data collection method
which is taken only before and after a meal. As a patient
needs to measure his/her glucose level at any time of the
day, the relationship developed from the data which
considers all the factors rather than conventional one might
improve the accuracy of the data.10 selected data are
presented in the TABLE-I.
Fig. 6: Clarke Error Grid Analysis (Without Notch Filter)
TABLE I. SELECTED TRAINING DATA Later considering the effect of 50Hz ac line frequency
Patient ADC Value Glucometer Value
we designed an RC low pass filter and added the filter
1 851 97 across the voltage divider output. Then we collected data
2 889 125 from this modified circuit and analyzed the data in Matlab.
3 868 105 We developed another equation for new dataset.
4 840 88 y=0.0799*x-564.9 (3)
5 870 114
6 850 91
Here, y=glucose level, x= adc value
7 874 124 The plot of ADC value vs. glucose concentration for the
8 851 96 new dataset is shown in Fig. 7. We also took test data for
9 860 100 our modified circuit. Then we compared the data with a
10 850 88 commercial glucometer.

Then we plotted the ADC value of the data against glucose

concentration determined from the commercial invasive
device. After regression analysis in Matlab, we developed
an equation which we used later to determine the blood
glucose level by using our device.
Equation developed from the training dataset is,
y=0.9893*x-747.9 (2)
Here, y=glucose level, x=adc value

Fig. 5 shows the plot of ADC value vs. glucose level for
training data. For the testing purpose of the proposed
prototype, we took data from 5 people. We determined
blood glucose level by our device and compared the value
with the commercial invasive glucometer value. For error
analysis, we used Clarke Grid Error (CGE) analysis. Fig. 7: ADC value vs. Glucose Value (With Notch Filter)
It was found that the test measurements are within the Clarke Error Grid analysis for filtered data is shown in Fig.-
accepted region. Fig. 6 shows the Clarke grid analysis for 8.
the measured data.
 96

[7] T. W. Jones, et al., "Mild hypoglycemia and impairment of brain stem

and cortical evoked potentials in healthy subjects," Diabetes, vol. 39, pp.
1550-1555, December 1, 1990.
[8] S. Clarke and J. Foster, "A history of blood glucose meters and their
role in self-monitoring of diabetes mellitus," British Journal of biomedical
science, vol. 69, p. 83, 2012
[9] D. C. Klonoff, “Continuous glucose monitoring: Roadmap for 21st.
century diabetes therapy,” Diabetes Care, vol. 28, pp. 1231–1239, May
2005
[10] Sparacino, G., Facchinetti, A. & Cobelli, C. “Smart” continuous
glucose monitoring sensors: on-line signal processing issues. Sensors 10,
6751–6772 (2010).
[11] Yadav, Jyoti & Rani, Asha & Singh, Vijander & Murari, Bhaskar.
(2015). Prospects and limitations of non-invasive blood glucose monitoring
using near-infrared spectroscopy. Biomedical Signal Processing and
Control. 18. 10.1016/j.bspc.2015.01.05
[12] Harman-Boehm, I., Gal, A., Raykhman, A. M., Naidis, E. & Mayzel,
Y. Noninvasive glucose monitoring: increasing accuracy by a combination
of multi-technology and multi-sensors. Journal of diabetes science and
technology 4, 583–595 (2010)
[13] K. Yan, D. Zhang, D. Wu, H. Wei and G. Lu, "Design of a Breath
Fig. 8: Clarke Error Grid Analysis for filtered test data Analysis System for Diabetes Screening and Blood Glucose Level
Though test data lies in the acceptable region of the Prediction," in IEEE Transactions on Biomedical Engineering, vol. 61, no.
Clarke grid analysis, it contains error. The reason behind the 11, pp. 2787-2795, Nov. 2014
error might have occurred as our proposed model does not [14] Martín-Mateos, P. et al. In-vivo, non-invasive detection of
hyperglycemic states in animal models using mm-wave spectroscopy. Sci.
calculate the effect of IR that comes from sunlight. Also, the Rep. 6, 34035
error might occur as the intensity of light reaching the [15] Tamada, J. A. et al. Noninvasive glucose monitoring - Comprehensive
photodetector changes slightly with time. Regression clinical results. Jama-J Am Med Assoc 282, 1839–1844
analysis may also introduce a slight bit of error as the [16] https://www.thescientist.com/?articles.view/articleNo/50631/title/Will-
the-Noninvasive-Glucose-Monitoring-Revolution-Ever-Arrive-/ (6/6/2018)
dataset is not large. The total cost of our device is BDT [17] O.S. Khalil, Non-invasive glucose measurement technologies: an
1200 or $14 which is lower than commercially available update from 1999 to the Dawn of the new millennium, Diabetes Technol.
devices. From the aforementioned discussion, it can be said Ther. 6 (5) (2004) 660–697
[18] N. Peled, D. Wong, S.L. Gwalani, Comparison of glucose levels in
that our proposed device costs lower than any other
capillary blood samples obtained from a variety of body sites, Diabetes
implemented devices mentioned in our literature review. Technol. Ther. 4 (1) (2002) 35–44
Also, our device is much simpler than other implemented [19] M.R. Robinson, R.P. Eaton, D.M. Haaland, G.W. Keep, E.V. Thomas,
devices. B.R. Stalled, P.L. Robinson, Non-invasive glucose monitoring in diabetic
V. Conclusion patients: a preliminary evaluation, Clin. Chem. 38 (1992) 1618–1622
[20] K. Maruo et al., "Noninvasive blood glucose assay using a newly
developed near-infrared system," in IEEE Journal of Selected Topics in
In this paper, we have developed a near-infrared LED- Quantum Electronics, vol. 9, no. 2, pp. 322-330, March-April 2003.
based non-invasive blood glucometer which uses diffuse [21] Heinemann, L., Schmelzeisen-Redeker, G. & on behalf of the Non-
reflectance to measure the blood glucose level. The invasive task force (NITF) Diabetologia (1998) 41: 848.
[22] S. Haxha and J. Jhoja, "Optical Based Noninvasive Glucose
prototype is of low cost ($14) which fulfills our primary Monitoring Sensor Prototype," in IEEE Photonics Journal, vol. 8, no. 6, pp.
object to make the low-cost device. The simplicity of the 1-11, Dec. 2016.
prototype can be used to convert the prototype into a [23] J. Yadav, A. Rani, V. Singh, and B. M. Murari, "Near-infrared LED-
based non-invasive blood glucose sensor," 2014 International Conference
wearable device. In future errors present in our work can be
on Signal Processing and Integrated Networks (SPIN), Noida, 2014, pp.
reduced by considering the effect of sunlight and make the 591-594
light reaching the photodetector more stable. Also, machine [24] Zhanxiao Geng, Fei Tang, Yadong Ding, Shuzhe Li & Xiaohao
learning algorithms can be used to make the regression Wang, “Noninvasive Continuous Glucose Monitoring Using a
analysis more accurate. Though our device does not give the MultisensorBased Glucometer and Time Series Analysis” , Scientific
Reports 7, Article number: 12650 (2017)
highest accuracy, its simplicity and low cost can have a vital [25] A. Tura, A. Maran, G. Pacini, Non-invasive glucose monitoring:
impact on wearable glucometer technology and continuous assessment of technologies and devices according to quantitative criteria,
glucose monitoring system. Diabetes Res. Clin. Pract. 77 (2007) 16–40.
[26] www.its.caltech.edu/~bi177/private/L6_handout.pdf
REFERENCES [27] Kim, B. M. (2004). Optical Property Measurements of Turbid Media
Using Continuous-Wave Light Sources. Journal of the Korean Physical
[1] S Coster, MC Gulliford, PT Seed, JK Powrie, R Swaminathan, Society, 44(2), 427-434
“Monitoring blood glucose control in diabetes mellitus: a systematic [28] Y. Ozaki, Near-infrared spectroscopy – its versatility in analytical
review”, Health Technology Assessment, vol. 4, no.12, 2000 chemistry, Anal. Chem. 28 (6) (2012) 545–563
[2] Guariguata, L. et al. Global estimates of diabetes prevalence for 2013 [29] B. Rabinovitch, W.F. March, R.L. Adams, Non-invasive glucose
and projections for 2035 for the IDF Diabetes Atlas. Diabetes Research and monitoring of the aqueous humor of the eye. Part I. Measurement of very
Clinical Practice, 137–149 (2013). small optical rotations, Diabetes Care 5 (1982) 254–258
[3] Zhang, M., Xu, W. & Deng, Y. A New Strategy for Early Diagnosis of [30] A.K. Amerov, J. Chen, W. Small Gary, M.A. Arnold, Scattering and
Type 2 Diabetes Mellitus by standard-free, label-free LC-MS/ MS absorption effects in the determination of glucose in whole blood by near-
quantification of glycated peptides. Diabetes, DB_130347 (2013) infrared spectroscopy, Anal. Chem. 77 (14) (2005) 4587–4594.
[4] IDF, http://diabetesatlas.org/resources/2017-atlas.html (6/6/2018) [31] O.S. Khalil, Spectroscopic and clinical aspects of
[5] D. E. Trachtenbarg, "Diabetic ketoacidosis," American Family noninvasive glucose measurements, Clin. Chem. 45 (2) (1999)
Physician, vol. 71, pp. 1705-1714, 2005. 165–177.
[6] “Identification of Hypoglycemia and Hyperglycemia in Type 1
Diabetic Patients Using ECG Parameters” Linh Lan Nguyen, Steven Su,
Member, IEEE, and Hung T. Nguyen, Senior Member, IEEE