Satuluri V. K. R. Rajeswari1
Ponnusamy Vijayakumar1*
-
(Department of Electronics and Communication Engineering, SRM Institute of Science
and Technology/Kattankulathur-Chennai, Tamilnadu-603203, India {rs3740, vijayakp}@srmist.edu.in)
Copyright © The Institute of Electronics and Information Engineers(IEIE)
Keywords
Artificial intelligence, Diabetes mellitus, Deep learning, NIR
1. Introduction
Diabetes Mellitus(DM) is a chronic disease with a growing incidence worldwide.DM is
collectively used for all types of diabetes. According to the diabetes atlas, one
in 11 adults (20-79 years of age) have diabetes (463 million people); one in two adults
with diabetes is undiagnosed(232 million people); one in five people with diabetes
is above 65 years of age (136 million people). More than 1.1 million children and
adolescents below 20 years have type 1 diabetes. Three out of four (79%) people are
living in low middle-income countries, one in six live births (20 Million) are affected
by hyperglycemia in pregnancy, while 84% of them have gestational diabetes, and 10%
of global health is spent on diabetes [1].
Individuals with diabetes are always at a higher risk of surgical and nosocomial infections.
The possibility of developing cancer is often found in Type-2 DM. Infections due to
yeast, fungi, and bacteria are found more in people with DM [2]. It has become crucial to have real-time insight into the glucose levels in an individual.
There is a strong need for a self-care management device for the daily monitoring
of glucose.
This study evaluated different types of glucose monitoring and how non-invasive glucose
monitoring is effective in daily life. Artificial intelligence-powered algorithms
that contribute to the novel cause of non-invasive glucose monitoring are also discussed.
1.1 Classification of Diabetes Mellitus
Diabetes Mellitus is a collective term used for different types of diabetic conditions.
Diabetes is classified into Type-I, Type-II, and Gestational diabetes.
1. Type-I Diabetes Mellitus:
Type-I diabetes is commonly found in people under the age of 35 years . Due to genetic
influences and other environmental factors. It is also known as juvenile diabetes
because it occurs during adolescence [3].
2. Type-II Diabetes Mellitus:
Type-II DM occurs as a result of the insufficient production and secretion of insulin.
It is the major type of diabetes worldwide, and the incidence increases with age [3].
3. Gestational Diabetes Mellitus:
Gestational DM occurs during pregnancy. There are considerable fluctuations in glucose
levels during this period, which sometimes leads to other complications, even after
the end of the gestational period [3]. Although there are different types of DM, a medium to monitor glucose should be
considered of the utmost importance. The following section examines different ways
of glucose monitoring.
2. Glucose Monitoring
There are diagnostic criteria for monitoring glucose in DM. An individual is categorized
as diabetic when the upper limit exceeds the described limits [3]:
Random plasma glucose ${\geq}$200mg/dL(11.1mmol/L);
Fasting plasma glucose${\geq}$126 mg/dL(7mmol/L);
Oral glucose tolerance test>200mg/dL(11.1 mmol/L).
The tests to analyze the blood glucose concentrations are generally taken from a pathology
lab. Blood is drawn from the arm, and the glucose concentrations are determined from
the blood. Furthermore, different types of monitoring glucose will be elaborated on
and discussed.
2.1 Invasive, Non-invasive Monitoring, Minimally Invasive Glucose Monitoring
The glucose concentration is determined by harvesting blood from the hand(venepuncture
method). The glucose concentrations are tested from the plasma or serum. This is a
general method followed in hospitals for checking the diabetic condition in patients.
Continuous blood withdrawal from a patient under ICU care or surgical procedure has
a potential risk of a decrease in blood levels. Tests performed at the laboratories
can sometimes show false results due to toxicity and cross-reactions due to various
agents [4]. Due to the rise in diabetes for a decade, it has become vital to have a continuously
monitored glucose device. Different methods for glucose monitoring will be discussed
in the following subsection. The glucose concentration can be detected by invasive,
minimally invasive, and non-invasive methods [5], as depicted in (Fig. 1). In the invasive method of detecting glucose, blood is drawn from the body. Apart
from laboratory tests, which are not suited for continuous monitoring, there are many
products in the market that allow continuous monitoring of glucose invasively. A glucometer
is a device consisting of a lancet that helps to prick the finger for capillary blood.
The glucose strip present reacts and detects the amount of glucose by oxidizing the
amount of blood and producing a current proportional to the glucose level. Through
the meter, an electron travels to the voltage to the current converter and produces
a voltage proportional to the glucose. The drawback of this approach is that pricking
the finger for every reading is quite painful, and the lancets and test strips are
expensive [5].
There are minimally invasive (MI) techniques that need a minimum intervention of the
inner body for the analysis. This could be the upper layers of skin, interstitial
fluid, or tears that must be extracted to measure the glucose concentration [5] (Fig. 2) shows the different minimally invasive methods.
In the ultrasound approach of the MI method, interstitial fluid is sampled using ultrasound
methods, and surface plasma resonance is detected. The refractive index is measured
using microsystem technology, which gives the glucose concentration. The drawback
of this approach is that ultrasound is susceptible to temperature [5,6]. The reverse iontophoresis approach of MI method depends on the circulation of a
small electric current between the anode and cathode that is located on the surface
of the skin. This approach allows access to the small interstitial fluid. A current
is produced when sodium ions migrate by causing the convective flow of the fluid.
This fluid also carries glucose molecules that reach the cathode. The glucose sensor
at the cathode measures the glucose concentration by following the oxidation process
(enzymatic method). The drawback of this method is that it is susceptible to sweat
due to heat generation, and skin irritation can occur due to the passage of current.
Quick changes in measuring the glucose current are not possible [5,7].
The minimally invasive method of sonophoresis measures glucose by extracting the interstitial
fluid using an enzymatic method. This method uses low-frequency pressure waves to
trigger glucose waves through the skin [5]. This method is used more commonly for drug delivery than for detecting glucose [8]. There may be erroneous readings due to interference from other compounds and pressure
changes. This method could also produce incorrect readings due to temperature variations
[5,8].
The Non-Invasive method of detecting glucose are divided into thermal, electric, and
optical methods, as depicted in Fig. 3. The thermal method of detecting glucose uses the heat generation method. Photoacoustic
spectroscopy(PAS), metabolic heat conformation(MHC), and thermal emission spectroscopy
(TES) are the three types of thermal methods of detecting glucose. In electric methods
of detecting glucose, the dielectric properties of glucose at low frequencies are
detected. This is achieved using small amounts of electromagnetic radiation, current,
and radiation. Bioimpedance spectroscopy, electro-magnetic sensing, and millimeter
and microwave sensing are the three electric non-invasive approaches. The third methodology,
optical methods, uses light that penetrates the skin or sample at a specific wavelength
and detects glucose. The different types of noninvasive optical methods include terahertz
time-domain spectroscopy, Raman spectroscopy, FIR spectroscopy(far-infrared spectroscopy),
MIR spectroscopy(mid-infrared spectro-scopy), NIR spectroscopy(near-infrared spectroscopy),
optical coherence tomography(OCT), and optical polarimetry [5].
Among the available market solutions and extensive research on invasive, minimally
invasive, and non-invasive methods, extensive research is occurring in the non-invasive
monitoring of glucose. The following drawbacks of invasive and minimally invasive
techniques justify the need for a non-invasive methodology for detecting glucose [5,9]:
1. Anxiety is always triggered when pricking the finger and the pain associated with
it. For continuous monitoring of glucose, there is a need to prick the finger continuously,
which will not work.
2. For continuous monitoring of glucose, it is essential for a system to be small
so that it can be carried anywhere.
3. Lancets and test strips are only used once. They are quite expensive.
4. There is a possibility of incorrect recordings caused by environmental changes.
5. Incorrect readings can occur if the testing strip is misplaced.
6. Erroneous readings are unavoidable when the finger is not cleaned when the test
is taken.
7. In a minimally invasive method, removing the interstitial fluid is painful. This
method is not recommended for the continuous monitoring of glucose.
8. In a minimally invasive method, sensors are inserted inside the human body and
send important information to IoT devices. This method is associated with allergies
and emergencies.
The above drawbacks justify the need for the non-invasive monitoring of glucose. Former
sections have shown that non-invasive methods can solve the above-discussed challenges.
The next section evaluates non-invasive methods for detecting glucose using different
approaches.
Fig. 1. Glucose detection approaches.
Fig. 2. Minimally invasive approach to detecting glucose.
Fig. 3. Non-invasive approach of detecting glucose.
Fig. 4. Categorization the ML algorithms.
3. Non-invasive Glucose Monitoring
The World Health Organization (W.H.O) reported a higher number of deaths due to strokes,
ischemia, and diseases, such as blindness, kidney failure, heart attacks, and lower
limb amputation associated with diabetes [10]. There is a need to monitor glucose continuously for emergency conditions, before
and after surgery, after release from ICU, at home, and to monitor the chronic condition
itself. Few studies have worked on different non-invasive glucose monitoring methods,
as shown in Fig. 3. Different approaches towards detecting glucose non-invasively are discussed further.
In the thermal method of detecting glucose, PAS uses short laser pulses that implement
wavelengths absorbed by any molecule and produce microscopic localized heating. It
is susceptible to noise and temperature changes [11-13]. By implementing NIR wavelengths, glucose is detected at 1540-1840nm [11] and 905 nm-1550nm [12]. By implementing MIR in an in-vivo method of detecting glucose from the skin, glucose
molecules are detected at 1070,1105, and 1140cm$^{-1}$ [13]. This approach fails in the in vivo method because of the poor sensitivity of glucose
detection.
In the thermal method of MHC, physiological parameters in the form of radiation, convection,
and evaporation are measured using the metabolic oxidation of glucose [14,15].Glucose is detected from the amount of heat dissipated. Thermal and optical sensors
are generally used to measure the physiological parameters, such as blood flow rate
[14,15] of local tissue, blood oxygen [14], hemoglobin, and oxyhemoglobin concentrations [15]. The susceptibility to temperature and sweat are the main drawbacks of this approach.
In the TES thermal method of detecting glucose, the heat emitted by the human body
is in the form of energy, which can be detected at the FIR range of 8 ${\mathrm{\mu}}$m
to 14${\mathrm{\mu}}$m.Glucose molecules in the body absorb this radiation and provide
a detection mechanism for glucose. A handheld approach to detecting glucose was presented
in [16], where the device was made from a filter spectrometer and the rmobile detector. While
sensitive to temperature and heat, tissue density varies for each person, which may
produce fault readings. Continuous glucose monitoring is not possible, which is another
drawback of this approach.
In the electrical method of bioimpedance sensing, the permittivity and conductivity
of the membrane in red blood cells produce dielectric impedance when current is applied.
Bioimpedance strip electrodes are placed on the wrist to collect glucose measurements
[17]. As this procedure may produce fluctuated data due to temperature sensitivity and
physiological conditions that can affect the cell membrane, another approach for obtaining
better accuracy was proposed [18], where MIR was also implemented. The glucose levels are detected at (5-10)${\mathrm{\mu}}$m
at 100 Hz-30MHz. Sensitivity to water is another drawback for both approaches. In
the electrical method of the microwave/ millimeter detection of glucose, radiation
is present in a low energy per photon and less scattering in the tissue, which is
advantageous for detecting glucose concentration much more accurately. A radar-driven
microwave sensor is proposed in [19] to detect glycemia levels in patients with diabetes. The transmission of mm waves
to detect glucose from two microstrip patch antennas is proposed in [20]. This method is used widely in communications that need detection through the exploitation
of transmission and absorbance characteristics owing to the advantages of deeper penetration.
This approach faces challenges when there is a variation in blood molecules, and physiological
parameters, including breathing and temperature changes in the body.
There are many approaches to the optical method of measuring glucose molecules. Terahertz
or FIR spectroscopy is based on the principle of absorption due to the vibration and
rotation of weak and heavy bonds of atoms. This is achieved at wavelengths between
0.3THz (1000${\mathrm{\mu}}$m)and 30THz(10${\mathrm{\mu}}$m). The silicon Dowe prism
for internal reflection was used in [21]. Glucose molecules were detected at 0.1-0.5THz, and similarly at 0.1-1.0THz [22]. When the blood glucose levels rise above the normal value, the amplitude and phase
of the reflection coefficient change on human skin. If there is a strong absorption
of water, identification of glucose molecules becomes difficult, which is a drawback
of this approach. Raman spectroscopy is the principle of determining the scattered
light from a monochromatic light. Miniaturized Raman spectroscopy was designed from
the benchtop [23]. Glucose molecules were detected at 1125cm$^{-1}$.The collection time for glucose
in this method was large, which is a drawback. In [24], ellipsoidal reflector-based spectrometry was proposed to detect glucose. Glucose
was also detected at 980nm while not using reflector spectrometry to increase the
accuracy. The interferences in blood with other molecules, such as hemoglobin, challenge
this approach. Another disadvantage of this approach is the predictability of the
wavelength. In the MIR optical method, glucose can be detected at 12-30THz. Owing
to the longer wavelength, there is light scattering, but the water molecules cannot
be absorbed in MIR ,resulting in less penetration. Hence lasers, such as quantum cascade
lasers, were proposed by [25] with this approach. An external cavity QC laser between 8-10${\mathrm{\mu}}$m was
focused on a hollow core fiber to deliver light. Glucose molecules are detected at
1080 cm$^{-1}$. Similarly, an ATR prism was used in [26]. Two FTIR spectrometers were used to detect the glucose. Three different wavenumbers,
i.e, at 1050 cm$^{-1}$, 1070 cm$^{-1}$, and 1100 cm$^{-1}$, were chosen to detect
glucose, but glucose peaks were obtained at 1036 cm$^{-1}$,1080 cm$^{-1}$, and 1100
cm$^{-1}$.The prototypes designed in [25,26] are expensive, which is a challenge in continuous glucose monitoring. The NIR approach
to detecting glucose is a wider area of research for designing wearables [27,28]. Glucose can be detected between 780 nm to 2500nm.Using a PPG circuit and NIT amplifier,
glucose is detected at on a single wavelength, 940nm in [27]. This method has the drawback of accuracy, which is much enhanced in [28], where glucose is detected in the range of 940nm by reflection spectroscopy and at
1300 nm by detection spectroscopy. Owing to the low wavelength, there is high light
scattering, which is a challenge in NIR spectroscopy.
Optical coherence tomography is an imaging method of detecting optical characteristics
in bio-tissues at micrometer resolutions. The method works on the principle of low-coherence
interferometry and coherent radiation. The blood glucose level is obtained at the
dermis layer of the skin using this approach. A Michelson interferometer, fiberoptic
coupler, and broadband light source for illumination are gathered to detect glucose
[29]. Glucose is detected by measuring the exponential slope of the light attenuation
in the tissue. This approach is sensitive to skin thickness, temperature changes,
and tissue homogeneity. Optical polarimetry works on the principle of chiral molecules
that can rotate on the plane of polarized light. Glucose is a chiral molecule when
it rotates in the polarization plane of a light beam by an angle ' ${\alpha}$' in
a clockwise direction. The rotation is proportional to the concentration of glucose,
optical path length, temperature, and wavelength of the laser beam. Glucose is detected
between 635nm and 830 nm. This method is quite sensitive to interference, temperature,
and motion changes because the optical rotation is minimal, and there is high scattering
of light [30].
This section described the different approaches for monitoring glucose non-invasively.
These approaches have been adapted to achieve sensitivity, coefficient of determination,
and accuracy. Methods that provide the best accuracy and reliability for a system
to work with the least error are vital in healthcare. The following section discusses
the ML algorithms implemented for the different approaches to detecting glucose non-invasively.
4. Integration of IoT and Machine Learning for Monitoring Blood Glucose Non-invasively
AI is the discipline of making computers learn human intelligence. ML is a part of
AI that helps train computers to learn.AI applications in healthcare ranging from
detection, prediction, diagnosis, self-management, and personalization of disease
therapy [31].
The three categories of ML algorithms are supervised learning, unsupervised learning,
and reinforcement learning. In supervised learning, machines are trained using 'labeled'
data. Classification and reinforcement learning are two categories of supervised learning.
In unsupervised learning, machines are trained using 'unlabeled' data.
The two learning classifications of unsupervised learning are association learning
and clustering learning. The third classification in ML is reinforcement learning
which is implemented to make decisions. For detection, ML techniques from different
studies are gathered and discussed further .For a non-invasive diabetic management
system to thrive, understanding proper models in ML and choosing suitable algorithms
for the nature of data is very important. The glucose levels in blood (non-invasively)
were predicted, and many approaches of ML were explored.
Every ML algorithm follows the stages from collecting the data/datasets, data pre-processing,
feature selection/ extraction, detection, and diagnosis of glucose monitoring [31].
4.1 Discussion on ML Techniques in Diagnosing Diabetes/Glucose Monitoring by Integrating
IoT
Data collection/Datasets: In healthcare scenarios and applications, ML algorithms
are first evaluated on publicly available datasets, such asthePIMA dataset and Kaggle.
The public dataset on non-invasive monitoring of glucose is unavailable. This paper
described the number of volunteers whose glucose levels were tested while implementing
different ML approaches. The accuracy of the model depends on the quality and quantity
of data [31].
1. Data Pre-Processing: Once the data is collected, the next step is to clean and
process it. The data are processed using different techniques depending on the data
gathered and the nature of the system [31]. The nature of data collected in [32, 33, 37, 38] is the PPG signal that comprises noise, artifacts, and baseline drifts. These are
reduced using Digital wavelet transform(DWT), Wavelet decomposition, and Moving average(MA)
[32]. The activity detection module in [33] extracted cleaner data without noise, whereas filters were implemented in [37,38] to remove unwanted observations. The pulse waveforms(PSW) data in [34,35] are processed polynomial fitting for removing baseline drift [34] and customized modeling to remove noise from raw optical signals. The spectroscopy
data in [36, 41,42] were filtered using filters, such as row normalization, butter worth filter [36], and Kalman filter [42].
2. Feature Selection: The efficiency of an algorithm depends on the selection and
extraction of features from a dataset. It helps in decreasing the complexity and time
taken to process, decreasing overfitting while increasing accuracy [31]. The nature of data, e.g., image, text, texture, color, and shape, depends on the
modality used in the application. The features selected are glucose molecules in optical
NIR spectroscopy [27, 36, 37, 39-42]. The features extracted in NIR spectroscopy are the amplitude, difference of optical
density, variance, skewness, and standard deviation [32]. The pulse waveform from arteries was collected; the distinct feature extracted was
the amplitude [34]. By implementing optical granular computing, the low and high pressure, rising phase,
and temperature are extracted from the pressure signal [35]. The blood flow rate is extracted from the PPG signal and blood oxygen saturation
[38]. The above data, it can be inferred that for glucose molecules, the body signals
are common features that may provide extraordinary results for research in the detection
and diagnosis of th DM.
3. Detection and diagnosis of glucose monitoring: In this study, many researchers
have explored and implemented various approaches of ML when surveying for the best
papers while integrating the IoT and ML.
In applications of NIR spectroscopy of detecting glucose molecules, a deep neural
network(DNN) with R=0.97,Mean absolute relative difference(mARD)= 4.86,Average error(AvgE)=4.88,Mean
absolute deviation (MAD)=9.42,Relative mean square error(RMSE)=13.57 is achieved [27]. In another study implementing NIR spectroscopy, partial least square (PLS) was used
,where the average correlation coefficient Rp is 0.86 [32]. An artificial neural network (ANN) [36] is implemented. ANNs perform better with a Clarke-error grid=86% compared to bioimpedance
spectroscopy and Alternating least square(ALS). The blood oxygen and blood flow rate
is extracted from the PPG signal to analyze the glucose concentration non-invasively.
The coefficient of determination (R$^{2}$=0.91), Clarke error grid =83% in class A
and 17% in class B is obtained [38]. ANN is applied to build a model on glucose molecules extracted by breathing acetone.
A low mean square error, Regression=1, is achieved. Neural network (NN) performs better
than time and frequency, and single pulse analysis showed regression=1 and an error
limit of ${\pm}$7.5% [39]. The DNN, ANN, and NN are deep neural networks with advantages in dealing with intra
class differences and noisy information. NN consists of enormous numbers of neurons
and is advantageous in DM, where the data are massive. A supervised ML Support vector
machine (SVM) is implemented on the extracted glucose levels and amplitude. It helps
classify the nonlinear data to linear data [34,41]. This alters the DM training data in a higher dimension. The root mean square error(RMSE),
which depicts the difference between actual outcome and predictions, achieved RMSE=19.90mg/dL
[41] when compared with the autoregressive integrated moving average (ARIMA), random forest
(RF). The RF is the next commonly implemented ML algorithm for regression and classification
problems. Compared to SVM, ridge linear regression (RLR), multilayer neuron perceptron
network (MNPN) RF has a better coefficient of determination (R\-$^{2}$ =0.90) and
a Clarke error grid of 87.7% [33]. The sweat sensor is designed to extract glucose levels from sweat. Linear regression(LR)
and polynomial regression (PR)analysis is applied to building the model [37,40]. Correlation factors P=0.672,0.574, and 0.343 were achieved, whereas R$^{2}$=0.851
was achieved [37]. On a nonlinear dataset, PR was implemented, which produced a correlation coefficient
of R$^{2}$=0.99 [40]. A customized physique-based fuzzy granular modeling(PbFG) achieved R$^{2}$=0.9 and
Clarke(A+B)>90% when compared with PbFG SVR and PbFG ANN [35]. In a similar modality of detecting blood glucose non-invasively, SVM with the linear
kernel is applied and showed a classification of 80.7% [34].
Different approaches toward the prediction and diagnosis of non-invasive methods of
glucose monitoring were elaborated. The following section discusses the justification
of the suitable methodology for detecting glucose from blood non-invasively.
Table 1. Literature review on modality and ML algorithms implemented for a non-invasive method of monitoring glucose.
|
Ref
|
Modality
|
ML
Algorithm
Applied
|
Performance Metrics
|
|
[27]
|
NIR spectroscopy
|
DNN
|
R=0.97
mARD=4.86
AvgE)=4.88
MAD)=9.42
(RMSE)=13.57
|
|
[32]
|
NIR spectroscopy
|
PLS
|
Rp=0.86
|
|
[33]
|
PPG waveform
|
RF
|
R=0.90
Clarke(A+B)=87.7%
|
|
[34]
|
Pulse waveforms
|
SVM-linear kernel
|
Classification of diabetic and pulse waveform=80.7%
|
|
[35]
|
Pulse waveforms
|
Fuzzy granular modeling
|
R$^{2}$=0.851
Clarke(A+B)>97.9%
|
|
[36]
|
Bioimpedance spectroscopy
|
ANN
|
Clarke-error grid=86%
|
|
[37]
|
Electrochemical method
|
LR
|
P=0.672,0.574,0.343
R$^{2}$=0.95
|
|
[38]
|
Single pulse analysis
|
ALS
|
R$^{2}$=0.91,
Clarke error grid =83% in class A and 17% in class B
|
|
[39]
|
Biosensor
|
ANN
|
Regression=1
Error limit=±7.5
|
|
[40]
|
Electrochemical impedance spectroscopy
|
PR
|
R$^{2}$=0.99
|
|
[41]
|
Time series forecasting
|
RF
|
RMSE=19.90mg/dL
|
5. Challenges and Future Works
Researchers have focused on developing desktop models and larger equipment by integrating
IoT and ML to design a non-invasive model for detecting glucose. There is research
potential in designing a portable model. A non-invasive glucose-monitoring device
should provide real-time measurements irrespective of interference and noise. The
data collected from the device must channel toward a cloud or any database to store
the measurements, transfer them to healthcare providers, and generate alerts during
an emergency. The predictions from the data can also make future doctor visits much
easier to access and understandable and provide a better quality of treatment. The
predictions should be modeled to help the doctors go through the patient's history,
analyze the disease status with predictions, and provide better quality treatment.
Wearables proposed in [27, 32, 37, 41, 42] detect glucose levels and make predictions but do not generate alerts or enhancements.
Researchers can consider this as future work. Future work can be done on the ML algorithms
applied in [32-35, 37, 40-42] that can be replicated with the different DL algorithms and improve the accuracy
of the proposed models. Future work can also be done by considering more features,
i.e., heartbeat, blood oxygen level, and blood pressure, in the developed model. Correlation
analysis of the blood sugar with additional features can be taken as future work.
A wearable that can detect blood glucose levels by providing accurate real-time predictions
with built-in features of the heartbeat pattern, blood oxygen levels, and pulse rate
can improve healthcare.
6. Conclusion
This paper presented the integration of IoT and ML in diagnosing non-invasiveglucose
monitoring. This paper elaborated on the various approaches todetect glucose molecules
from blood non-invasively. Furthermore, this paper presents comprehensive research
activities on different ML algorithms that address the detection anddiagnosis of glucose
issues non-invasively. On the other hand, much potential for designing portable, detection,
predictions, alerts, and other enhancements with correlation analysis with additional
features can be addressed in future work.
REFERENCES
International Diabetes Federation , 2019, Diabetes Atlas, Key Findings, 2019, 9th
ed., Vol. 2019
Szablewski L., Sulima A., 2017 Apr 1, The structural and functional changes of blood
cells and molecular components in diabetes mellitus., 398, Vol. 4, pp. 411-423
Kaul K, Tarr JM, Ahmad SI, Kohner EM, Chibber R., 2012, Introduction to diabetes mellitus.,
771, pp. 1-11
Principles and Problems of Blood Glucose Measurement
Villena Gonzales W, Mobashsher AT, Abbosh A., 2019 Feb 15, The Progress of Glucose
Monitoring-A Review of Invasive to Minimally and Non-Invasive Techniques, Devices
and Sensors. Sensors (Basel), Vol. 19, No. 4, pp. 800
Li D., et al. , 2006, A Novel Minimally Invasive Method to Detect Glucose Concentration
without Blood Extraction, 2006 1st IEEE International Conference on Nano/Micro Engineered
and Molecular Systems, pp. 1185-1189
Yao Yao, Chen Jingyao, Guo Yuhan, Lv Tian, Chen Zilin, Li Ning, Cao Shaokui, Chen
Bingdi, ,Integration of interstitial fluid extraction and glucose detection in one
device for wearable non-invasive blood glucose sensors,Biosensors and Bioelectronics,
Volume 179,2021,113078, ISSN 0956-5663,
Mitragotri S., Kost J., 2000 May-Jun, Low-frequency sonophoresis: a noninvasive method
of drug delivery and diagnostics, Biotechnol Prog, Vol. 16, No. 3, pp. 488-92
2 Dec. 2019, The Drawbacks of Invasive Glucometer., GLUCOACTIVE
WHO. , 13 Apr. 2021, Diabetes., World Health Organization: WHO
Dasa Manoj Kumar, Markos Christos, Janting Jakob, Bang Ole, 2019, Multispectral photoacoustic
sensing for accurate glucose monitoring using a supercontinuum laser, Journal of the
Optical Society of America B, Vol. 36, No. A61
Pai Praful, Sanki Pradyut, De Arijit, Banerjee Swapna, 2015, NIR Photoacoustic Spectroscopy
for Non-Invasive Glucose Measurement, Conference proceedings: Annual International
Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering
in Medicine and Biology Society. Conference. 2015
Sim JY, Ahn CG, Jeong EJ, Kim BK., 2018 Jan 18;, In vivo Microscopic Photoacoustic
Spectroscopy for Non-Invasive Glucose Monitoring Invulnerable to Skin Secretion Products,
Sci Rep., Vol. 8, No. 1, pp. 1059
Fei Tang, Wang Xiaohao, Dongsheng Wang, Junfeng Li, 2008, Non-Invasive Glucose Measurement
by Use of Metabolic Heat Conformation Method, Sensors, Vol. 8
Cho OK, Kim YO, Mitsumaki H, Kuwa K., 2004 Oct, Noninvasive measurement of glucose
by metabolic heat conformation method, Clin Chem, Vol. 50, No. 10, pp. 1894-8
Carl D Malchoff , Kamal Shoukri , Julian I Landau , Jansuz M Buchert , 1 Dec. 2002,
A Novel Noninvasive Blood Glucose Monitor, Diabetes Care, Vol. 25, No. 12, pp. 2268-2275
Pathirage KD , Roopasinghe P , Sooriyaarachchi JJ , Weththasinghe R , Nanayakkara
ND. , 2019 Jul, Removing subject dependencies on Non-Invasive Blood Glucose Measurement
using Hybrid Techniques, Annu Int Conf IEEE Eng Med Biol Soc, Vol. 2019, pp. 7197-7200
Carlos F Amaral , Martin Brishwein , Bernhard wolf , June 2009, Multiparameter Techniques
for Non-Invasive Measurement of Blood Glucose, Sensors and Actuators B: Chemical,
Vol. 140, No. 1, pp. 12-16
Omer A. E., Shaker G., Safavi-Naeini S., May 2020, Portable Radar-Driven Microwave
Sensor for Intermittent Glucose Levels Monitoring, in IEEE Sensors Letters, Vol. 4,
No. 5, pp. 1-4
Saha S., et al. , 2016, Evaluation of the sensitivity of transmission measurements
at millimeter waves using patch antennas for non-invasive glucose sensing, 2016 10th
European Conference on Antennas and Propagation (EuCAP), pp. 1-4
Cherkasova O. P., Nazarov M. M., Shkurinov A. P., 2016, Application of terahertz time-domain
spectroscopy for blood glucose monitoring, 2016 International Conference Laser Optics
(LO), pp. S2-11-S2-11
Gusev S., Guseva V., Simonova Anna, Demchenko Petr, Sedykh Egor, Cherkasova Olga,
Khodzitsky M., 2017, Application of terahertz pulsed spectroscopy for the development
of non-invasive glucose measuring method, pp. 3229-3232
Sang Park Y , Ahn S , Chang H , Lee W , Hyun Nam S. , 2020 Jul, Influence of Raman
Spectrometer Collection Efficiency on Performance of Noninvasive Blood Glucose Detection
for Device Miniaturization, Annu Int Conf IEEE Eng Med Biol Soc, Vol. 2020, pp. 6139-6142
M. Bezuglyi , N. Bezuglaya , O. Kuprii , I. Yakovenko , 2018, The non-invasive optical
glucometer prototype with ellipsoidal reflectors, 2018 IEEE 59th International Scientific
Conference on Power and Electrical Engineering of Riga Technical University (RTUCON),
pp. 1-4
Liakat S, Bors KA, Xu L, Woods CM, Doyle J, Gmachl CF, 2014 Jun 23, Noninvasive in
vivo glucose sensing on human subjects using mid-infrared light, Biomed Opt Express,
Vol. 5, No. 7, pp. 2397-404
Kasahara R., Kino S., Soyama S., Matsuura Y., 2018, Noninvasive glucose monitoring
using mid-infrared absorption spectroscopy based on a few wavenumbers, Biomed. Opt.
Express, Vol. 9, pp. 289-302
Hina A., Saadeh W., June 2020, A Noninvasive Glucose Monitoring SoC Based on Single
Wavelength Photoplethysmography, in IEEE Transactions on Biomedical Circuits and Systems,
Vol. 14, No. 3, pp. 504-515
Joshi A. M., Jain P., Mohanty S. P., Agrawal N., Nov. 2020, iGLU 2.0: A New Wearable
for Accurate Non-Invasive Continuous Serum Glucose Measurement in IoMT Framework,
in IEEE Transactions on Consumer Electronics, Vol. 66, No. 4, pp. 327-335
Amrane S., Azami N., Elboulqe Y., 2015, Optimized algorithm of dermis detection for
glucose blood monitoring based on optical coherence tomography, 2015 10th International
Conference on Intelligent Systems: Theories and Applications (SITA), pp. 1-5
Malik BH, Coté GL., 2010 Jan-Feb;, Real-time, closed-loop dual-wavelength optical
polarimetry for glucose monitoring, J Biomed Opt, Vol. 15, No. 1, pp. 017002
Jyotismita Chaki , S. Thillai Ganesh , S.K Cidham , S. Ananda Theertan , Machine learning
and artificial intelligence based Diabetes Mellitus detection and self-management:
A systematic review, Journal of King Saud University - Computer and Information Sciences,
ISSN 1319-1578
Vega Pradana Rachim , Wan-Young Chung , 2019, Wearable-band type visible-near infrared
optical biosensor for non-invasive blood glucose monitoring, Sensors and Actuators
B: Chemical, ISSN 0925-4005, Vol. 286, pp. 173-180
Enric Monte-Moreno , 2011, Non-invasive estimate of blood glucose and blood pressure
from a photoplethysmo-graph by means of machine learning techniques, Artificial Intelligence
in Medicine, ISSN 0933-3657, Vol. 53, No. 2, pp. 127-138
Chun Ouyang , Peng Zhou , Zhongxue Gan , 2020, A non-invasive test method for type-2
diabetes mellitus by pulse waveform fitting, Biomedical Signal Processing and Control,
ISSN 1746-8094, Vol. 60, pp. 102000
Weijie Liu , Anpeng Huang , Ping Wang , Chao-Hisen Chu , 2019, PbFG: Physique-based
fuzzy granular modeling for non-invasive blood glucose monitoring, Information Sciences,
ISSN 0020-0255, Vol. 497, pp. 56-76
Tronstad C., Elvebakk O., Staal O. M., Kalvøy H., Høgetveit J. O., Jenssen T. G.,
Birkeland K. I., Martinsen Ø. G., 2019, Non-invasive prediction of blood glucose trends
during hypoglycemia, Analytica chimica acta, Vol. 1052, pp. 37-48
Hong Yongseok Joseph , Hyunjae Lee , Jaemin Kim , Minha Lee , Hyung Jin Choi , Taeghwan
Hyeon , Dae Hyeong Kim , 10 Oct. 2018, Multifunctional Wearable System That Integrates
Sweat-Based Sensing and Vital-Sign Monitoring to Estimate Pre-/Post-Exercise Glucose
Levels, Advanced Functional Materials, Vol. 28, No. 47, pp. 1805754
Habbu Shraddha , Dale M , Ghongade R. , Estimation of Blood Glucose by Non-Invasive
Method Using Photoplethysmography, Accessed 11 June 2021
Thati Anand, Biswas Arunangshu, RoyChowdhury Shubhajit, Kumar Tapan, 2015, Breath
acetone-based non-invasive detection of blood glucose levels, International Journal
on Smart Sensing and Intelligent Systems, Vol. 8
Bhide A., Muthukumar S., Prasad S., 2018, CLASP (Continuous lifestyle awareness through
sweat platform): A novel sensor for simultaneous detection of alcohol and glucose
from passive perspired sweat, Biosensors & bioelectronics, Vol. 117, pp. 537-545
Rodríguez-Rodríguez I., Rodríguez J. V., Chatzigiannakis I., Zamora Izquierdo M. Á.,
2019, On the Possibility of Predicting Glycaemia 'On the Fly' with Constrained IoT
Devices in Type 1 Diabetes Mellitus Patients, Sensors (Basel, Switzerland), Vol. 19,
No. 20, pp. 4538
Amine Rghioui , Assia Naja , Jaime Lloret Mauri , Abedlmajid Oumnad , Jan. 2021, An
IoT Based Diabetic Patient Monitoring System Using Machine Learning and Node MCU,
Journal of Physics: Conference Series, Vol. 1743, pp. 012035
Author
Vijayakumar Ponnusamy received his Ph.D. from SRM IST in 2018. His area of research
was Applied Machine Learning in Wireless Communication (cognitive radio). He completed
his Masters in Applied Electronics from the College of Engineering, Guindy in 2006.
In 2000, he received his B.E in Electronics and Communication Engineering from Madras
University. He is currently working as a Professor in the ECE Department, SRM IST,
Chennai, Tamil Nadu, India. He is a Certified “IoT specialist” and “Data scientist.“
He is also a recipient of the NI India Academic award for excellence in research (2015).
His current research interests are Machine and Deep learning, IoT-based intelligent
system design, Blockchain technology, and cognitive radio networks. He has authored
or co-authored more than 100 International journals and more than 65 International,
National conferences. He is a senior member of IEEE.
S. V. K. R. Rajeswari is a Research Scholar in SRMIST, Kattankulathur, Chennai-Tamil
Nadu, India. She is pursuing her Ph.D. at the faculty of Electronics and Communication
Engi-neering department. She received her M.Tech with Embedded Systems Technology
as her specialization. and Bachelor of Technology from J.N.T.U.H, India. Her research
interests include data science, machine learning algorithms, deep learning, artificial
intelligence, IoT, and embedded systems.