A sensor based intelligent system for classification and assistance of diabetes patients in telemedicine technology

Abstract

Diabetes is one of the chronic metabolic disorder. Under diabetic condition, blood glucose level should be properly maintained in order to avoid various major diseases. The condition will be worse when it is not controlled at an earlier stage. Even massive heart attack cannot be identified when the patient has been affected by diabetes. Early diagnosis is required for preventing fatal diseases like cardiac problem, asthma, heart attack etc. In the proposed system measurement of glucose level and Prediction/ diagnosis of diabetes is based on the real time low complexity neural network implemented on a wearable device. A larger network is required for the diagnosis which needs to be present far-off in cloud and initiated for diagnosis and classification process of diabetes whenever it is essential. People can be able to manage and monitor the required basic parameters like heart rate, glucose level, lung condition, pressure of blood using the corresponding light weight biosensors in the wearable device designed through telemedicine technology. The quality of the disease diagnosis and Prediction is improved in this way. Using neural network feed forward prediction model in conjugation with back propagation algorithm and given training data, the system predicts whether the patient is prone to diabetes or not. The proposed work was evaluated using physic sensor data from physio net data base and also tested for real time functioning. The Proposed system found to be efficient in accuracy, sensitivity and fast operative.

Keywords

Continuous glucose monitoring diabetic condition prediction early medical diagnosis ANN remote health assistance

1 Introduction

A metabolic disorder, also known as diabetes is characterized by either insufficient or ineffective insulin and high blood glucose. Diabetes may lead to blindness, renal failure, amputation, heart attack and stroke. In many developed countries it is considered as the third leading cause of death. In 2010, there were 285 million people who were suffering from diabetes globally. In the absence of better control or cure, it is estimated that this count may increase to 430 million. At present, diabetes is considered as a kind of the most costly, restricting, and deadly ailments determined in several of the countries. At an alarming rate, this disease continues to increase. It is estimated that 9.6 million women are having diabetes. In diabetes, people either suffer from shortage of insulin or has a decreased ability to utilize their insulin. DM abbreviated as Diabetes Mellitus is a long-lasting and advanced metabolic disorder. Around millions of people in the world suffering from diabetes, according to the World Health Organization (WHO). It is predicted that the number of diabetic patients can extend by quite 100% by the year 2030 [1]. Insufficient insulin production by pancreas, ineffective use of the insulin produced by the pancreas or hyperglycaemia are the common manifestations of diabetes. Obesity, hypertension, elevated cholesterol level, high fat diet and sedentary lifestyle are the common factors of diabetes. Improper management and late diagnosis of diabetes are followed by development of kidney failure, blindness, other kidney diseases and heart related diseases. There is no established cure for diabetes. But, by using popularly known treatments, correct diet and proper exercises, the glucose level of affected patients can be managed.

Diabetes can be categorized as three subgroups. They are, type 1 diabetes (T1D), type 2 diabetes (T2D), and gestational diabetes (GDM). The normal effects of insulin are restricted to T2D patients and they eventually give up the ability to generate sufficient amount of this hormone. For patients with T2D, there is a broad category of healing procedures followed in medical field. They commonly receive medications at the primary phase of disease. This shows improvement in production and utilization of insulin. But they must receive external doses of insulin eventually. T1D patients suffer intense loss in the secretion of insulin on the other hand, and need to take insulin as external medication completely to keep control of their blood glucose. Couple of types of T1D treatment is available. Standard quantities of insulin through Multiple Daily Injections as per prescription or constant subcutaneous insulin infusion (CSII) with the help of a pump.

GDM can be kept I control just like T2D. But due to the interaction between insulin and hormones released by the placenta, it only occurs during pregnancy. Timely diagnosis, education of patients in self-management and continuous medical care are required in each class of diabetes to prevent severe complications (e.g., ketoacidosis) and minimize the risk of long-term complications (e.g., nephropathy, retinopathy, diabetic foot, cardiovascular disease, or stroke) [2]. Apart from medication, controlling diabetes needs few more self-care habits. They are, scheduling meals carefully, less intake of carbohydrates, workout, keep watching BG levels regularly, and correcting activities periodically. The effects of non-adherence to recommended treatment are not immediately evident. The effects of lasting problems develop only after years.

Many systems were developed to assist diabetic patients for early prediction and proper treatment. But, each of them has their own disadvantage or any performance parameter like accuracy, specificity, speed of operation, reliability, robustness lagging which stands as a challenge for their efficiency. The proposed system and methodologies used shows good improvement with respect to these parameters comparatively, hopefully assisting and guiding the users to take necessary treatment by means of prediction.

2 Related works

There are plenty of different injection regimens and varieties of insulin are available for diabetes. The acceptability for patients and the glucose readings that are obtained with the use of single or multiple-dose injection regimens. Even though the availability of insulin vials and pens this is not to the desired level. The same quality of service is not provided by hospitals even though they provide the same type of service. There is no existence of previous analysis that is used to identify which data mining technique can provide more reliable accuracy. It takes more time.

A similar work projects a way supported the classification of polygenic disorder that uses AdaBoost algorithmic rule with call stump as base classifier for classification [3]. Later for any accuracy verification Support Vector Machine, Naive mathematician & call Tree area unit used as base classifiers for AdaBoost algorithmic rule. The accuracy is obtained from AdaBoost algorithmic rule with call stump as base classifier is eighty. 72% that is larger once compared to Support Vector Machine, Naive mathematician & call Tree. For helping doctors, call network area unit introduced victimization varied data processing algorithms. The main objective is to established associate data system which provides higher accuracy for classifying the sort of polygenic disorder, which takes place in four phases. Initial part is dataset assortment, second part embody coaching of the world dataset with AdaBoost algorithmic rule, third part includes confirming of native dataset & final part includes accuracy once boosting with error nineteen. 27%, which is lower in comparison to all or any different classifier. SVM doesn’t shows any improvement in accuracy (79.68%) & most accuracy was shown by call stump seventy-four to eighty.72%

Another recent method for information analysis ways to research the biomarkers in breath for diseases detection that uses feature extraction & classification algorithms. This method explores the employment of one-dimensional(1-D) convolution neural network (CNN) algorithmic rule [4]. This, mixes feature extraction & classification technique obtained from associative array of gas sensors. During this paper, breath signals needed for information associate analysis is measured victimization an array of metal chemical compound semiconductor gas device. at first options area unit extracted from raw analog signals & applicable classifiers area unit accustomed segregate the gas concentrator as traditional or abnormal. options like magnitude, difference, derivative, integral, time options area unit calculated & Support Vector Regression algorithms were accustomed observe polygenic disorder. Feature extraction techniques like Principal part algorithmic rule (PCA), Linear Discriminant Analysis (LDA), & classification algorithmic rule like call tree, K- nearest neighborhood, ANN, perceptron, SVM area unit usually used algorithmic rule. Thus, the algorithmic rule reduces the machine value & will classify raw breath signals victimization One-Dimensional CNN algorithmic rule & mean sq. error [7].

A Unique technique for designation of polygenic disorder that takes place in 2 stages to predict polygenic disorder standing [5]. The initial prediction stage involves 2 machine intelligence & data engineering techniques like formal logic (F), neural network (N) & case base reasoning (C) as a personal approach (FCN). Then the ultimate prediction stage involves rule based mostly algorithmic rule for the values obtained from initial stage. So, this method benefitted for higher accuracy of prediction rate in comparison to older ways. The method undergoes through many steps for complete designation sort of a. Login page b. Collect, analyze & normalize the input parameters c. Applying neural network technique d. Applying fuzzy thought e. Applying CMB approach f. Applying rule based mostly algorithmic rule g. Diagnosis report. This paper includes combination of formal logic, neural network, & CMB approaches for higher designation of polygenic disorder. Thus, a replacement approach is projected referred to as FNC approach for diagnosis polygenic disorder by victimization recently designed influenced input parameters [8].

A related perspective work projected a system for polygenic disorder designation known as LDA–MWSVM [6]. They proposed a method to extract features and reducing victimization. The Linear Discriminant Analysis (LDA) methodology with categorization targeting the MWSVM type [7]. Gangji associated Abadeh projected a pismire Colony-based arrangement to abstract a collection of fuzzy rules, named FCS-ANTMINER, for polygenic disorder designation. In, authors restrained aldohexose prediction as a variable regression downside utilizing Support Vector Regression (SVR).

A different method intends to predict polygenic disorder maculopathy by using an information adjustive neuro fuzzy reasoning classifier [9]. The authors of [10] aim at the level of intensity in DR, employing a computer-aided screening system (DREAM) that evaluates biological structure representations with different brightness and fields of read via machine learning techniques. A 2-part methodology, Diabetic bodily structure Image convalescence (DFIR), was utilized in for DR estimation [10]. The primary part implements feature choice on digital retinal bodily structure pictures. The next part employs a support vector machine for predicting. A distinct side to the DR downside analyzed by [11] therein, a technique for calculating the necessity for consultation was developed, supporting the detection of DR-related lesions in pictures of retina. Atlast, Giancardo et al. projected a strategy for Diabetic macular dropsy forecasting, that may be a usual that challenges vision of DR.

A most recent technique employed machine learning ways in partially automated type labelled coaching sets for forming composition models [12]. In this work, the authors projected a fuzzy ontology-based, Case-based reasoning (CBR) framework, depicting knowledgeable reasoning and assessed on polygenic disorder designation issues. This method performs associate analysis of mining type of Stream Classifiers for on time Clinical call Support Systems.

A similar method proposed a system where the implementation and analysis are done by Both regression and classification algorithms, including training by static and dynamic methods [13]. The data set consists of 89 CGM time series restrained in diabetic patients for 7 continuous days. Evaluation of Performance is done by accessing and comparing the event-prediction abilities of the methods usage and scrutinized. The mathematical results proved that fair performance achieved by static techniques, whenever regression methods are studied. Anyhow, classifiers given good classification efficiency when trained for a specific event category, like hyperglycemia, attaining matchable efficiency with regressors, fetching the benefit of sooner event prediction.

A related work proposed wearable micro system. This system is particularly used for blood glucose monitoring which is minimally invasive, autonomous and pseudo-continuous [14]. This system addressed a growing demand for the replacement of tedious finger pricking tests for diabetic patients. This design takes a whole blood section from a minimally lanced skin wound using a SMA-based micro actuator. After that it measures the blood glucose level from the sample. In vitro characterization determined that the SMA micro actuator produced penetration force of 225 gf, penetration depth of 3.55 mm, and consumed approximately 5.56 mW·h for triggering [26]. The micro actuation mechanism was also evaluated from the blood samples that are obtained from the wrist off human volunteers. Its Blood glucose monitoring, but Real implementation is not possible.

A more relevant method proposed that for the development of a continuous non-invasive glucose monitoring system, near infrared photoacoustic spectroscopy can be utilized. By using field programmable gate array (FPGA), a portable embedded system is implemented for taking photoacoustic measurements on tissues [15]. This is to estimate the glucose concentration. This examining technique of body glucose is verified in vitro on glucose liquids and in vivo on tissues, with light and sound signal amplitude varying linearly with trial glucose application. To estimate glucose concentration from photoacoustic measurements, a kernel-based regression algorithm using multiple features of the photoacoustic signal can be used. It easy, safe and secure but the cost is high. The dataset consists of 89 CGM time series measured in diabetic subjects for 7 subsequent days. Performance metrics have been introduced and analyzed specifically to assess and compare the event prediction capabilities of the methods. But its high cost and the sensor values is not accuracy.

3 Proposed methodology

The proposed system consists of two separate units for measuring the blood glucose level and a unit to pump the insulin. The output value from the GMU is given to the microcontroller unit which will drive the insulin pumping unit. The method consists of IoT with hardware connection system development to facilitate the process of patient’s personal monitoring and diagnosis. By using Arduino board, the blood glucose level is read. Using wireless communication, glucose value level is logged to the system. Web page is interfaced with data collection [16]. The database receives the readings from Arduino via ESP 8266 Wi-Fi Module. This module can be accessed by the patients or the registered doctors. To independently monitor the patient’s diabetic health, this research is significant. Medial officers in the hospitals are directly alerted by the IoT system. This system also involves an ANN classification algorithm which is stored in the cloud to analyze the measured glucose level of patients to give predictions on their diabetic condition and classify the type of diabetes they are prone to get affected. Figure 3.1 shows the overall flow of the proposed work.

Fig. 3.1

Block representation of the proposed work.

Artificial neurons are available in ANNs. Every artificial neuron will have the following structure. A node, also called as ‘body’ is used for processing which is represented by circles. ‘dendrites’ are represented as ‘connections from other neurons’ and ‘axons’ are represented as ‘connections to other neurons’ (Fig. 3.2). In a commonly used ANN architecture, the neurons are arranged in layers. An ordered set of predictor variables (a vector) is used to represent the input layer. The distributed value from the input layer neurons is received by the intermediate layer neurons. For each input an intermediate neuron there exists a connection weight in-between. The input neuron and connection weight got multiplied and given to the intermediate middle neuron. The sum of the weighted inputs of each neuron in the middle layer is calculated. Then applies a non-linear operation to the sum. The output of that particular middle neuron is obtained from the result of the function. The output neuron is connected with each middle neuron. There exists a connection weight for every link between an intermediate neuron and the output neuron. As a last procedure, the output neuron receives the weighted total of inputs. Then finally puts on the non-linear function to the summated weight [17]. The output for the entire ANN is the result of this function.

Fig. 3.2

Architecture of ANN for diabetic prediction and classification.

The layers can be categorized as,

(a) Input layer

It is the responsible layer for the retrieval of information, signals, features or measurements from the external environment. All these inputs are normalized within the limit values produced by activation functions. The result of this normalization is better numerical precision for the mathematical operations that are performed by the network [13].

(b) Hidden, intermediate, or invisible layers

For extracting patterns that are associated with the process or system being analyzed these layers are made by the neurons. For performing the most of the internal processes from a network these layers are used.

(c) Output layer

This layer is responsible for producing and presenting the final network outputs. This is the result from the processing performed by the neurons in the previous layers.

4 Hardware, data acquisition and training of algorithm

4.1 Data acquisition

The data is provided to a continuous monitoring system which has a connection of neural network in sensors. The system measures the body condition based on different parameters like breath testing, monitoring body temperature and glucose level [18]. Initially the raw data signal in the form of analog signal is sent to glucose sensor which sends the modified analog signal microcontroller.

4.1.1 Analog to analog transmission

Lower bandwidth having bandpass characteristics are required by the analog transmission. This transmission is the simplest form of modulation. The analog signal which modulates the amplitude of the carrier wave is known as modulating signal. To modulate a carrier and the modulated signal, a signal is used [19]. The mathematical representation of the actual message or baseband signal is shown below, $em (t) = Em \cos (2 π fmt)$ (4.1)

The high frequency carrier wave is given by, $ec (t) = Ec \cos (2 π fct + Φ c) .$ (4.2)

The modulated signal’s mathematical expression can be given as $s (t) = (Ec + Em \cos 2 π fmt) \cos 2 π fct$ (4.3)

‘m’ which is known as the modulation index, can be written as

$\begin{matrix} m & = (Emax - Emin) / (Emax + Emin) \\ = Em / Ec, where Emax = Ec + Em \end{matrix}$ (4.4)

$\begin{matrix} Emin & = Ec - Em, and s (t) \\ = Ec (1 + m \cos 2 π fmt) \cos 2 π fct, \end{matrix}$ (4.5)

Modulated output signal’s envelope can be given as, $1 + m em (t), for m < 1$ (4.6)

Loss of information occurs when m > 1 [21]. Figures 4.1 and 4.2 depict the proposed ANN based diabetes classifier and the configuration of ADC for arduino respectively.

Fig. 4.1

Flow chart of the proposes ANN diabetes classifier.

Fig. 4.2

Configuration of ADC for arduino.

4.1.2 Analog to digital transmission

To calculate the sign-to-noise ratio (SNR) we can use the gain block’s input. This work presents a methodical approach to the design of a gain block and ADC combination. We need to pick an ADC that has a better signal-to-noise ratio. To calculate the effective number of bits SNR will be helpful. The following equations shows this relationship. Any good ADC data sheet always specifies both SNR and bits [20].

Later the converted digital signal is received through sensors are trained using microcontroller and sent to a WIFI module ‘ESP8266’ is 1 MiB of built-in flash, allowing for single-chip devices capable of connecting to Wi-Fi and also contain UART (Universal Asynchronous Receiver-Transmitter) (on dedicated pins, plus a transmit-only UART can be enabled on GPIO2).

4.2 Training of ANN with back propagation algorithm

In artificial neural networks, backpropagation method is used to calculate a gradient. A gradient is needed to calculate the weights that are used in the network. This is the common method used for training. The BPNN is the type of Artificial Neural Network. A n number of input and one output back propagation system has m back propagation $IF x_{1} is A_{i}^{k} and \dots and x_{n} is A_{n}^{k} THEN y is B^{k},$ where x_i is the input back propagation linguistic variable and y is the output back propagation linguistic variable [22]. Back propagation morphological standards A_i^k_, ^Bk can be given by back propagation membership functions like given below, $μ_{A_{k}^{i}} {(x}_{i}) = \exp [- (\frac{x_{i} - a_{i}^{k}}{σ_{i}^{k}})^{2}]$ (4.7) ${mu}_{B^{k}} (y) = \exp [- (\frac{y - b^{k}}{η^{k}})^{2}]$ (4.8)

The reasoning of simple back propagation for n-input-1-output back propagation neural network is defined below: ${f (x}_{1}, \dots {, x}_{n}) = \frac{\sum_{k = 1}^{m} bk [_{i = 1}^{Π^{n}} μ_{A_{i}^{k} (x_{i})}]}{\sum_{k = 1}^{m} [_{i = 1}^{Π^{n}} μ_{A_{i}^{k} (x_{i})}]}$ (4.9)

The input data vectors in n- dimension can be given as x^p (i.e., x^p = (x₁^p, x₂^p, ... , x_n^p))and output data vector in1-D y ^p for p values 1,2,...,N, (that is N number of training sets of values). Energy function for p can be given as $E^{P} = \frac{1}{2} [{f (x}_{1}^{p}, \dots, x_{n}^{p}) - y^{p}]^{2}$ (4.10)

For simplicity, let E and f^p denote E^P and ${f (x}_{1}^{p}, \dots, x_{n}^{p})$ , respectively. After training the centers of output membership functions ( $\frac{\partial E^{P}}{\partial b^{k}}$ ), the widths of output membership functions ( $\frac{\partial E^{P}}{\partial σ^{k}}$ ), the centers of input membership functions ( $\frac{\partial E^{P}}{\partial a^{k}}$ ) and the centers of input membership functions ( $\frac{\partial E^{P}}{\partial σ^{k}}$ ), then we obtain the training algorithm: $b^{k} (t + 1) = b^{k} (t) - θ \frac{\partial E^{P}}{\partial b^{k}} |_{t}$ (4.11) $σ^{k} (t + 1) = σ^{k} (t) - θ \frac{\partial E^{P}}{\partial σ^{k}} |_{t}$ (4.12) $a^{k} (t + 1) = a^{k} (t) - θ \frac{\partial E^{P}}{\partial a^{k}} |_{t}$ (4.13) $η^{k} (t + 1) = η^{k} (t) - θ \frac{\partial E^{P}}{\partial η^{k}} |_{t}$ (4.14)

Where, η is the learning rate and t = 0,1,2, ...

ALGORITHM 1:

Given below are the steps for learning algorithms:

Phase 1: Provide the input data sample and calculate the corresponding output.

Phase 2: Compute the error rate that may exist among the output(s) and the estimated object(s).

Phase 3: Adjust the linking samples and membership relations as per the need.

Phase 4: Delete useless rule at a fixed number of epochs and membership function nodes, and add in new ones.

Phase 5: suppose error raised is greater than Tolerance, on that case go to Phase 1. Otherwise end the process.

Phase 6: END.

4.2.1 Experimental setup and real- time measurement

To develop a real time controller for automatic regulation of glucose levels, several ANN (Artificial Neural Networks) were employed in this work [24]. In Neural Network, blood glucose level prediction is based on glucose sensor. Figures 4.3 to 4.5 show the prototype of the complete experimental setup, glucose classification screenshot and lung sensor testing respectively.

Fig. 4.3

Prototype of the entire setup.

Fig. 4.4

Screenshot of glucose classification.

Fig. 4.5

Testing of lung sensor.

Testing blood sample: The blood glucose monitoring system has two aspects. They are blood sampling and glucose sensing. Developing a portable automated blood sampling system is the main aim of this work. This is for the integration of glucose sensor and, it can also be used in critical care settings.

Blood sampling automation minimizes excessive labour. It also reduces the contamination risk which is associated with repetitive sampling.

The customizable options that are used for sampling intervals lessens the need for caregiver intervention.

The minimization of blood sample obviates the need for volume replacement per analytical measurement.

Glucose sensing: For integrating with the sampling system, blood glucose sensing technologies have been evaluated. The system includes testing the blood sample using glucose biosensor which has a neural network consisting of neurons which senses the glucose level and classifies the type of diabetes [23]. The dataset covers the wide range of Glucose Infusion Rate (GIR), body temperature as well as environmental conditions (between 5–22μl/min). Similarly, by using a test set of biological data another two types of performance tests were carried out.

Firstly, the breath level is measured according to raising and decreasing of breath using lung sensor

Second body temperature is compared to the room temperature using temperature sensor.

Testing Lung Sensor: Using the lung sensor the breath level of the human body is evaluated to find whether the human body is affected with disease or not. The applications of breath sensors are abundant in disease diagnosis and high in scope.

5 Results and performance evaluation

A ROC curve (receiver operating characteristic curve) gives the graphical depiction of performance of a classifier used for prediction at all thresholding of classification. The performance of the classifier that is developed in this work is evaluated by calculating the accuracy, sensitivity and specificity [24]. Table 5.1 shows the threshold values of glucose value for type 1 and type 2 diabetes before and after food. These parameters are estimated as follows.

Table 5.1
Chart for diabetes classification based on glucose level

NORMAL TYPE 1 TYPE 2

Glucose Before Food After Food 126–146 ≥147

Level 0–90 0–126

	NORMAL		TYPE 1	TYPE 2
Glucose	Before Food	After Food	126–146	≥147
Level	0–90	0–126

The above Fig. 5.1 shows the ROC curve of accuracy of the classification used in the proposed system. The following. Table 5.2 shows how accuracy is evaluated with respect to other methods which are known to be good in accuracy.

Fig. 5.1

Graph for result accuracy.

Table 5.2

Accuracy Comparison of Classifiers

	AdaBoost algorithmic rule with call stump	Support Vector Machine	LDA–MWSVM	Proposed Method
Test 1	95.4	81	96.1	99.1
Test 2	93.6	80.2	98.7	99.5
Test 3	94.5	78.8	98.2	99.3
Test 4	93.9	79.4	97.9	99.2
Test 5	96.1	79.6	98.5	99.4

The following Fig. 5.2. shows the accuracy comparison graph for various similar classifiers.

Fig. 5.2

Accuracy comparison.

The below Table 5.3 contains the experimental results that are obtained from the proposed work. The performance evaluation is discussed for each person having Diabetic and Non-Diabetic conditions. This evaluation is done for a set of 300 patients.

Table 5.3

Performance metrics as comparison with other methods

Type used in method	Sensitivity	Specificity	Accuracy	TPR	TNR
Our designed model	89.10	86.8	99.1	86.64	88.92
SVM	87.14	82.4	81	82.31	86.53
AdaBoost algorithmic rule with call stump	81.03	79	95.4	85.72	84.98
LDA–MWSVM	80.45	78.4	96.1	81.4	82.2

Fig. 5.3. shows the performance comparison chart with respect to various evaluation parameters for nearly similar effective classifier and found proposed one stands ahead.

Fig. 5.3

Performance Evaluation graph.

6 Conclusion

The Intelligent technique proposed above will help in timely assistance for a patient suffering from chronic diseases like diabetes with the help of sensors used in telemedicine application [25]. For achieving good quality of service, the sensor used should be energy efficient and should have improved network life time by using optimal routing algorithm. The ANN algorithm used can be a back-propagation methodology to achieve more accuracy in classification. Glycemic variability is reduced and glycemic control is improved. Similarly, by using this method, severe hypoglycemia risk and the emergent medical needs are decreased. Also, the need for hospitalization, quality of life and satisfaction of treatment are improved. For uninterrupted continuous monitoring [26], the bulk data that is collected continuously need to be maintained in cloud which increases the size of cloud requirement and corresponding cost increase. This appears to be the major challenging factor of the proposed system. As future scope closer match with physiologic needs is allowed by programmable insulin delivery. Short- or rapid-acting insulin need to be used, which minimizes the peaks and absorption-related variability.

References

Yin

and Han

, CPAR: Classification based on predictive association rules. In Proceedings of the 2003 SIAM International Conference on Data Mining (pp. 331–335). Society for Industrial and Applied Mathematics. (2003).

Poonguzhali

and Chakravarthi

, Retinal Health Investigation by Segmentation of Major and Minor Blood Vessels in Fundus Images for Diabetes Patients, Journal of Clinical & Diagnostic Research 13(10) (2019). DOI: 10.7860/JCDR/2019/41221.13207

Vijayan

V.V.

and Anjali

, Decision support systems for predicting diabetes mellitus—A review. In 2015 Global Conference on Communication Technologies (GCCT) (2015), 98–103. DOI: 10.1109/GCCT.2015.7342631

Lekha

and Suchetha

, Real-time non-invasive detection and classification of diabetes using modified convolution neural network, IEEE Journal of Biomedical and Health Informatics 22(5) (2017), 1630–1636. DOI: 10.1109/JBHI.2017.2757510

Thirugnanam

, Thirugnanam

and Mangayarkarasi

, An ontology based system for predicting disease using SWRL rules, International Journal of Computer Science and Business Informatics 7(1) (2013).

Çalisir

and Dogantekin

, An automatic diabetes diagnosis system based on LDA-Wavelet Support Vector Machine Classifier, Expert Systems with Applications 38(7) (2011), 8311–8315. DOI:https://doi.org/10.1016/j.eswa.2011.01.017

Jitpakdee

, Aimmanee

and Uyyanonvara

, Asurvey on hemorrhage detection in diabetic retinopathy retinal images. In 2012 9th International Conference on Electrical Engineering/Electronics, Computer, Telecommunications and Information Technology (2012), (pp. 1–4).

Meigs

J.B.

, Nathan

D.M.

, D’Agostino

R.B.

and Wilson

P.W.

, Fasting and postchallenge glycemia and cardiovascular disease risk: the Framingham Offspring Study, Diabetes Care 25(10) (2002), 1845–1850.

Chamberlain

, Baird

, Dirani

and Guymer

, Unraveling a complex genetic disease: age-related macular degeneration, Survey of Ophthalmology 51(6) (2006), 576–586.

10.

Chowdhury

S.K.R.

, Smith

D.R.

and Fernyhough

, The role of aberrant mitochondrial bioenergetics in diabetic neuropathy, Neurobiology of Disease 51 (2013), 56–65.

11.

Pires

, Avila

, Jelinek

H.F.

, Wainer

, Valle

and Rocha

, Beyond lesion-based diabetic retinopathy: a direct approach for referral, IEEE Journal of Biomedical and Health Informatics 21(1) (2015), 193–200.

12.

Samant

and Agarwal

, Machine learning techniques for medical diagnosis of diabetes using iris images, Computer Methods and Programs in Biomedicine 157 (2018), 121–128.

13.

Gadaleta

, Facchinetti

, Grisan

and Rossi

, Prediction of adverse glycemic events from continuous glucose monitoring signal, IEEE Journal of Biomedical and Health Informatics 23(2) (2018), 650–659.

14.

Wang

A.A.

and Hutchinson

D.T.

, The effect of corticosteroid injection for trigger finger on blood glucose level in diabetic patients, The Journal of Hand Surgery 31(6) (2006), 979–981.

15.

Pai

P.P.

, Sanki

P.K.

, Sarangi

and Banerjee

, Modelling, verification, and calibration of a photoacoustics based continuous non-invasive blood glucose monitoring system, Review of Scientific Instruments 86(6) (2015), 064901.

16.

Borch-Johnsen

, Neil

, Balkau

, Larsen

, Nissinen

, Pekkanen...

and Jousilahti

, Glucose tolerance and mortality: comparison of WHO and American Diabetes Association diagnostic criteria, Lancet 354 (1999), 617–621.

17.

Montani

, Magni

, Bellazzi

, Larizza

, Roudsari

A.V.

and Carson

E.R.

, Integrating model-based decision support in a multi-modal reasoning system for managing type 1 diabetic patients, Artificial Intelligence in Medicine 29(1-2) (2003), 131–151.

18.

Schaller

, Lippert

, Schaupp

, Pieber

T.R.

, Schuppert

and Eissing

, Robust PBPK/PD-based model predictive control of blood glucose, IEEE Transactions on Biomedical Engineering 63(7) (2015), 1492–1504. DOI: 10.1109/TBME.2015.2497273

19.

Palmer

A.J.

, Weiss

, Sendi

P.P.

, Neeser

, Brandt

, Singh...

and Spinas

G.A.

, The cost-effectiveness of different management strategies for type I diabetes: a Swiss perspective, Diabetologia 43(1) (2000), 13–26.

20.

Valentine

W.J.

, Palmer

A.J.

, Nicklasson

, Cobden

and Roze

, Improving life expectancy and decreasing the incidence of complications associated with type 2 diabetes: a modelling study of HbA1c targets, International Journal of Clinical Practice 60(9) (2006), 1138–1145.

21.

Yan

and Zhang

, Feature selection and analysis on correlated gas sensor data with recursive feature elimination, Sensors and Actuators B: Chemical 212 (2015), 353–363.

22.

Priya

C.L.

, Gowthami

and Poonguzhali

, Lung pattern classification for interstitial lung diseases using an ANN-back propagation network. In 2017 International Conference on Communication and Signal Processing (ICCSP) (2017), (pp. 1917–1922). IEEE. doi:10.1109/ICCSP.2017.8286732

23.

Jacob

S.M.

, Raimond

and Kanmani

, Associated Machine Learning Techniques based On Diabetes Based Predictions, In 2019 International Conference on Intelligent Computing and Control Systems (ICCS) (2019), (pp. 1445–1450). IEEE., DOI: 10.1109/ICCS45141.2019.9065411

24.

Sivasangari

, Ajitha

, Rajkumar

and Poonguzhali

, Emotion recognition system for autism disordered people, Journal of Ambient Intelligence and Humanized Computing (2019), 1–7. https://doi.org/10.1007/s12652-019-01492-y.

25.

Poonguzhali

, Advanced fetal essential parameters observation using wsn, In International Journal of Pharmacy & Technology 8(4) (2016), 21629–21634.

26.

Wang

, Poscente

M.D.

, Park

S.S.

, Andrews

C.N.

, Pecht

O.Y.

, Mintchev

M.P.

, Wearable Microsystem for Minimally Invasive, Pseudo Continuous Blood Glucose Monitoring: The eMosquito, IEEE Transactions on Biomedical Circuits and Systems, (2017).