Design and development of mobile phone based … · Design and development of mobile phone based diabetes mellitus diagnosis system by using ANN - FP-growth techniques. K.Sridar1*,

Design and development of mobile phone based diabetes mellitus diagnosissystem by using ANN - FP-growth techniques.

K.Sridar1*, D. Shanthi2

1Department of Computer Science and Engineering, Odaiyappa College of Engineering and Technology, Theni, TamilNadu, India2Department of Computer Science and Engineering, PSNA College of Engineering and Technology, Dindigul, TamilNadu, India

Abstract

Diabetes mellitus is considered to be a chronic disease that makes millions of lives miserable. Type I andType II diabetes mellitus are caused depending upon the insulin level. Detection and diagnosis ofdiabetes mellitus has been performed effectively by various plasma glucose tests in the current medicalsystems including A1C test, fasting plasma glucose test (FPG) and oral glucose tolerance test (OGTT).Though the current diagnosis systems provide better results, the problems like imperfect concordancebetween the tests reduces the diagnosis accuracy. Another method of diagnosis is from the retinal imagesas diabetes causes retinal blindness, but these type of systems generally requires specialists to performoperation and also slightly expensive. Hence in this paper, the mobile based diabetes diagnosis system isproposed by obtaining the features from the real time inputs of glucometer with multiple measurementclinical data and retinal image features. The optimal features are selected using an optimizationtechnique while the statistical measures are calculated for the time series data features. These featuresare fused together and trained using neural networks called hierarchical extreme learning machines (H-ELM) NN for the generation of dataset with minimum error samples. Then the association rule miningalgorithm called FP-growth is employed to generate rules for determining the associations between thedifferent sets of data. Thus the type I and type II diagnosis mellitus is diagnosed effectively with betteraccuracy using the proposed mobile based diabetes diagnosis system.

Keywords: Diabetes mellitus, Plasma glucose test, Hierarchical extreme learning machines (H-ELM), FP-growth,Modified particle swarm optimization (MPSO).

Accepted on January 11, 2017

IntroductionDiabetes mellitus is a major metabolic disease in which theblood glucose level varies than normal over a prolongedperiod. There are three main types of diabetes mellitus. Type Iis caused due to the pancreas failure to produce sufficientinsulin. It is commonly known as insulin-depended diabetesmellitus as the disease is determined by considering the insulinlevel in the blood. It is caused due to the loss of beta cells inthe pancreas that produce insulin. Type I diabetes is of twokinds, immune-mediated and idiopathic. Type I diabetes ispartly due to inheritance, viral infections and environmentalfactors resulting in the death of beta cells but does not dependon lifestyle changes. Type II is commonly called as noninsulin-dependent diabetes mellitus is the most common andhighly affective diabetes. It is a condition in which the cells failto respond to insulin properly due to the defectiveresponsiveness of insulin receptor of the body tissues. Theinsulin sensitivity reduces continuously mainly due to thegenetics and lifestyle factors. The lack of physical activity,

excess fat, poor diet, stress and over dosage of sweeteneddrinks increases the risks of Type II diabetes. The third mostcommon diabetes is the gestational diabetes mellitus which iscommon in women during pregnancy that may improve ordecrease after delivery. It is quite similar to type II diabeteswith inadequate insulin secretion and low response of the bodycells.

Detection and diagnosis of diabetes is performed bydetermining the recurrent or persistent high blood glucoselevel. Fasting plasma glucose (FPG), plasma glucose, glycatedhemoglobin (A1C test), oral glucose tolerance test (OGTT),random plasma glucose test (RPGT) are some of the mosteffective tests for diagnosing diabetes mellitus. Diabetes isdiagnosed based on A1C criteria or the plasma glucose criteria,either the FPG value or the 2 h plasma glucose (2 h PG) valueafter the OGTT. Both detection and diagnosis is done by thesame tests for pre-diabetes in initial stages also. Though thismethod is costly and the non-availability of A1C testing in allregions, with incomplete correlation between A1C and average

ISSN 0970-938Xwww.biomedres.info

S73Special Section:Artificial Intelligent Techniques for Bio-Medical Signal Processing

Biomedical Research 2017; Special Issue: S73-S80

Biomed Res- India 2017 Special Issue

glucose level in certain individuals, the advantages makes it themost commonly implemented diabetes diagnosis test. A1C testconsiders factors such as age, race/ethnicity and anemia/hemoglobinopathies for the diagnosis. Additionally the FPGtest or the 2 h PG tests are utilized however the concordancebetween the A1C and FPG or 2 h PG tests and also betweenFPG and 2 h PG is imperfect. Hence the diagnosis accuracymay be affected.

A most common method of diabetes diagnosis, other thanplasma glucose tests, is the detection from the retinal images.The retinal images are generally captured using Funduscameras with microscopic lens and then the images are trainedwith the normal and infected retinal images to determine thediabetes. However this retinal image system is considerablyexpensive and it is also required to be processed by specialistsat the laboratories. Hence this method is not much possible forimplementation for the general public use. Replacing theexpensive components and improving the diagnosis test resultsby modifying the training algorithms and the data miningapproach can outperform the limitations.

In order to overcome the problems in the current diabetesdiagnosis systems, mobile-based diabetes mellitus diagnosissystem is proposed in this paper. The proposed diagnosissystem collects the real time data from the glucometer throughthe Bluetooth device in the mobile phones. The collected inputdata are transmitted to the operating system through Wi-Fi andare processed. The retinal images are captured using the mobilephone cameras and then are transmitted to the operating systemin a similar fashion. Likewise the multiple measurementclinical data are also input into the processing system. Thefeatures from these collected data are extracted and by using aModified Particle Swarm Optimization (MPSO), the optimalfeatures are selected. Similarly the statistical measures arecomputed for the time series data features and are fused withthe optimal features. Then the samples are trained using backpropagation NN but it has infinitesimal step size problem thatcannot be resolved in standard back propagation algorithm.Hence we use a special type of neural network calledHierarchical Extreme Learning Machine (H-ELM) NN inwhich the dataset is generated with selected minimum errorsamples. In order to determine the presence of diagnosis, theassociation rules are generated using Apriori algorithm but it isvery costly due to the candidate selection process. Hence weuse the FP-growth algorithm for generating association ruleswith which the relation between the features is estimated. Thusthe diabetes can be diagnosed accurately using the proposedmobile based diabetes diagnosis system.

The remainder of the paper is organized as follows: Section 2describes the previous research methods for the strategiesutilized in the diagnosis of diabetes that form the basicmotivation of our research. Section 3 explains the mobile-based diabetes diagnosis system with detailed illustration of theresearch methodologies. Section 4 presents the performanceevaluation results of the proposed diagnosis systems. Section 5concludes the research work and provides directions for thefuture research.

Related WorksBourouis et al. [1] presented a low cost and innovativedevelopment of Smartphone based intelligent system withmicroscopic lens which allows the patients in remote areas forregular disease diagnosis and eye examinations. The proposedwork used Artificial Neural Network algorithm for analyzingthe retinal images to identify the condition of the retinaldiseases. Karan et al. [2] proposed a novel approach for thediagnosis of diabetes with the help of neural networks andother computing technologies. The proposed work presentedthe initial results for a simple client and server two-tierarchitecture for healthcare. However it had highercomputational complexity. In Andrea et al. [3] the study of asmart phone enabling access to daily web based diaries andalso to the written feedback of an individual. The smart phoneswere used to register patients eating behavior to emotionsthrice a day by also registering their fasting blood glucoselevel.

Erkaymaz et al. [4] proposed a new approach for diagnosis ofdiabetes based on the Small-World Feed Forward ArtificialNeural Network (SW-FFANN). The classification performanceof the SW-FFANN was better than that of the conventionalFFANN. This work is high of cost and architecture.Roychowdhury et al. [5] proposed a computer-aided screeningsystem called DREAM where the fundus images with varyingillumination and fields of view were analyzed. The systemgenerated a severity grade for diabetic retinopathy (DR) usingmachine learning method. Classifiers such as Gaussian MixtureModel (GMM), k-nearest neighbor (kNN), support vectormachine (SVM), and AdaBoost were also analyzed forclassifying retinopathy lesions from normal ones. This systemwas tested on 1200 images from the publicly availableMESSIDOR data set that gave satisfactory results on theclassification. But the implemented system was not applicablefor data corresponding to increased coverage. Mohammad etal. [6] discussed on the Hypoglycemia which is one of the mostcommon neonatal disorders, associated with severecomplications. There has been a great deal of controversyregarding the definition and screening of hypoglycemia. A cut-off value for blood glucose level in glucometer readings wereproposed based on the study conducted on 238 newborns atrisk of hypoglycaemia.

Gaikwad et al. [7] presented a study was to compare bloodglucose in capillary finger-prick blood and gingival creviceblood using a self-monitoring blood glucose device amongpatients with gingivitis or periodontitis. Thirty patients withgingivitis or periodontitis and bleeding on probing (BOP) werechosen. The following clinical periodontal parameters werenoted: probing depth, BOP, gingival bleeding index, andperiodontal disease index. Bin-Sabbar et al. [8] proposed theimplementation of an integrated monitoring tool for thediabetes patients. The designed system provides a dailymonitoring and monthly services. The daily monitoringincludes recording the result of daily analysis and activates tobe transmitted from a patient’s mobile device to a centraldatabase. The result of this visit entered into the system and

Sridar/Shanthi

Special Section:Artificial Intelligent Techniques for Bio-Medical Signal ProcessingS74Biomed Res- India 2017 Special Issue

then synchronized with the central database. Finally, theendocrinologist can remotely monitor the patient record andadjust the treatment plan and the insulin doses if need. Singhalet al. [9] provide a medical image processing system based oncolorimetric detection particularly for diabetic patients that willmeasure the blood glucose levels by scanning an image of thevisual glucose test strip using a smart-phone.

Chikh et al. [10] discussed about Artificial ImmuneRecognition System (AIRS) was one of the methods that usedin medical classification problems where AIRS2 was anefficient version of the AIRS algorithm. The proposed workmodified AIRS2 called MAIRS2 where the K- nearestneighbors algorithm was replaced with the fuzzy K-nearestneighbors in order to improve the diagnostic accuracy ofdiabetes diseases. Bob Zhang et al. [11] proposed anoninvasive method to detect DM and non proliferativediabetic retinopathy (NPDR). Initially three groups of featureswere extracted from the images of retina. A color gamut wasestablished with 12 colors representing the features of theimages. 13 features were extracted from tongue images basedon measurements, distances, areas, and their ratios representthe geometry features. Applying a combination of the 34features, the proposed method can separate Healthy/DMtongues as well as NPDR/DM-sans NPDR (DM sampleswithout NPDR) tongues using features from each of the threegroups with average accuracies of 80.52% and 80.33%,respectively, but has high resistance to noisy images. Silva etal. [12] proposed a Bluetooth prototype low power battery-lesswireless sensor communication system that transmitsinformation to a mobile phone or windows PC application.Low Power Bluetooth was used in conjunction with harvestingRF methods which allowed the development of new wirelessbattery-less sensors may work for years without batteryreplacement.

Figure 1. Block diagram of proposed mobile-based diabetesdiagnosis system.

Mobile-Based Diabetes Diagnosis SystemIn this section, the development of the proposed mobile-baseddiabetes diagnosis system is illustrated in detail. The proposeddiagnosis system utilizes the features from the multiplemeasurements clinical data, real time input from theglucometer and the retinal image features. These features are

extracted using different approaches (Figure 1). The multiplemeasurements clinical data are collected from the patients fordifferent time periods in between a specified time interval.Different parameters are considered from the clinical data forthe diabetes diagnosis. The parameters are separatelydetermined for the Type I and Type II diabetes mellitus. Fromthe previous medical researches it is found that the factorsinfluencing the Type I and Type II diabetes are similar with thelifestyle behaviors being the only difference. Type II dependsupon the lifestyle changes while the Type I diabetes does notdepend upon it.

ParametersFrom the Dataset, the parameters are taken for type I and IIseparately. Type I diabetes mellitus considers the parameterssuch as age, gender, family history of diabetes, HemoglobinA1C (HbA1c), systolic blood pressure, platelet parameters,cholesterol, fast plasma glucose (FPG), impaired fastingglycaemia, impaired glucose tolerance, plasma glucose level,for the diagnosis. For diagnosing Type II diabetes mellitus, theparameters such as age, gender, cholesterol, smoking &drinking status, duration of illness, FPG, HbA1C, weightvariation, number of missing teeth, Papillary bleeding index,Russell’s periodontal index and community periodontalindex(CPI).

Data from glucometerThe inputs are also taken from the glucometer and also fromthe retinal images apart from the multiple measurementsclinical data. The glucometer is utilized for the home glucosemonitoring (HBGM) for patients whom are to be diagnosedwith diabetes mellitus. A small drop of blood obtained bypricking of skin in fingers and is placed on a disposable teststrip. The strip is read by the meter and uses it to calculate theblood glucose level which is displayed in the monitor in unitsof mg/dl or mmol/l. The readings from the glucometer arecollected and are fed into the operating system through amobile device. The glucometer test strips are made up ofchemicals like glucose oxidase and when the blood of 0.3 to 1μl is placed on the strip it reacts with the glucose oxidasewhich catalyzes the oxidation of glucose to gluconolactone.The data collected by the glucometer are transmitted to themobile phones through the Bluetooth and from the mobiles thedata is transmitted to the operating systems through Wi-Fi.

Retinal imagesThe retinal images are usually captured using the Funduscamera which captures high quality and high sensitive imageswith the microscopic lens. The image database is generated bycapturing images using the zoom lens mobile cameras so thatthe images can be captured at home itself and transmitted tothe operating systems. The images can also be taken fromSTARE database. The retinal images are pre-processed toremove the noise and various texture backgrounds. The greenchannel of the color retinal images is extracted as an RGBimage that provides higher contrast between vessels and

Design and development of mobile phone based diabetes mellitus diagnosis system


background. The approach includes the processes likebinarization, skeletonization and vessel thicknessmeasurement. Thus the features can be extracted based on thecategories. These features are also transmitted to the operatingsystems and are utilized in the diabetes diagnosis system.

Clinical dataMultiple measurements clinical data are collected from thehospitals treating the patients. The data quality is improved byusing data processing techniques like data cleaning, dataintegration, data transformation, and data reduction. These datawith unequal collecting frequency are processed beforeanalysis. A merging algorithm is utilized for combining thefeatures of multiple types observed at different times. Thus aset of clinical data that are related to diabetes are gatheredtogether.

Feature selection using modified PSOAs the elimination of the irrelevant features reduces thecomputational cost of the system, the optimal features areselected from the given set of input features includingglucometer features, retinal features and multiplemeasurements clinical data features. The best features areselected using modified PSO algorithm [13]. In modified PSOalgorithm, the group velocity of the particle swarm consideredin the normal PSO is converted into two separate and parallelparticle swarm velocity in order to improve the speed ofprocessing the data. Feature selection is a combinatorialoptimization problem in a discrete space. In modified PSO, theparticle velocity and position are updated using modifiedequations.�� + 1 = � *�� + �1 * �1 � * �� − �� + �2 * �2 � * �� − �� (1)�� + 1 = 10 �� < �� + 1��ℎ�� (2)Where Vij(k) is the jth dimensional speed in the kth particlesiteration, r1, r2 and rij are random numbers between 0 and 1, c1and c2 are the learning factors, w is the weighting factor, xij(k)is the jth dimension current position of particle i in the kth

iteration. pbetsij is the jth dimensional position of the individualextreme points in the particle i and gbetsij is the jth dimensionalposition of the global extreme points in the whole group.�� = 11 + exp −�� (3)In order to prevent the saturation of sig (Vij(k)), fewamendments are made. Then as stated the particle velocitygiven in equation (1) is converted into two velocities such that�� + 1 = �� + 1 + �� + 1 (4)As the amendments are made, the independent particle swarmsa and b does not have different speed limits so that these arethe two particle groups with different steps in the search space

in parallel optimization. The particle swarm with the largerstep size has strong global search ability and is faster in thesearch space optimization, while the particle swarm with thesmaller step size will have better local optimization ability. Inmodified PSO, to avoid the individual features pre-maturelyand then lower the optimization ability, the populationdiversity has to be guaranteed in the optimization. In order tomaintain two particle swarms in parallel, the particle position isexecuted as

�� + 1 = �� < �� ℎ�� (5)Where qj is the ratio of the number of j-dimension value (0 or1) in particle swarm to the number of all the particles while isthe convergence degree (normally 0.7-0.9).

The fitness function is given by� � = � �1 + � *� � (6)Where F(i) represents that he particle i generates the fitnessvalue of solution; p(i) represents that the feature subset uses themean of the correct classification rate; n(i) is the number ofselected features; is the weight parameter in the features. Thefeatures are selected in the following way

• Assume a j-dimensional target search space with d-particlescomposing a population

• First i-particles is expressed as a j-dimensional vectorXi=(xi1,xi2,…,xij)T; i=1,2,…,m

• If X’s ith bit value is 1, the feature is selected; else it is notselected

Algorithm 1: Modified PSO for feature selection

Size of groups = m;

Initialize particle swarm a, b

Initialize random position and velocity

Adjust velocity & position of the particle using equation (2)and related equations for (4)

Evaluate fitness of each particle using (6)

For particle i=1 to number of particles

If F(i)> fitness of pbestiUpdate pbesti = F(i)

For p ϵ neighborhood of F(i)

If F(p)>gbest then

Update gbest=F(p)

Next p

For each dimension j

Calculate particle velocity Vij(k)

If Vij(k) ≠ Vmax.Vmin then

Update particle velocity Vij(k)

Sridar/Shanthi


Update particle position xij(k)

Next dimension j

Next particle

Next generation until iteration stops

Statistical measuresAfter optimal feature selection, Statistical measures [14] arecalculated for explaining the data distribution in each period.As the selected features include time series features, they areneeded to be merged together. However the information suchas the tendency and the distribution of the features in theoriginal data may disappear when data merges. In order topreserve the information, the statistical measures arecomputed. The maximum and minimum descriptions areutilized for illustrating the extremes of the data. Mean is amethod for deriving the central tendency of a sample space,and standard deviation is a widely used measurement ofvariability or diversity. Variance is the measurement of contrastbetween the descriptions of relative features. Entropy is themeasurement of disorder in the distribution of the data features.Pearson’s correlation coefficient is a statistical technique thatshows whether and how strongly pairs of features are related,and express this relationship in values ranging between −1 and1. The closer the absolute value of the Pearson’s coefficient isto 1, the correlation between the features is the stronger.Statistical measures of disease related features are alsocomputed. The area of the disease spread is described. Meanintensity of the abnormality present in the diseased regions arecomputed. Using these statistical measures, the datadistribution over the period of time is expressed and the lostinformation is also highly retained. These statistical measuredfeatures are combined with the optimally selected featuresusing a merging technique.

Algorithm 2: Statistical measures

R=all Records before occurrence of event E

For each period t in feature subset

For each time-related feature F

F_Max t=maximum of all F in R within period t

F_Min t=minimum of all F in R within period t

F_Mean t=mean of all F in R within period t

F_SD t=standard deviation of all F in R within period t

F_Var t=variance of all F in R within period t

F_Ent t=entropy of all F in R within period t

F_Cor t=Pearson’s correlation of all F in R within period t

Add F_Max t, F_Min t, F_Mean t, F_SD t, F_Var t, F_Ent t,F_Cor t as additional features with the optimal features

End for

End for

Hierarchical extreme learning machine based neuralnetworks for selection of minimum error samplesAfter the selection and merge of features, the minimum errorsamples are to be selected to generate a dataset from which thematching process is performed to detect the diabetes mellitus.The features are computed using the artificial neural networksdue to the advantage of estimating most appropriate functionsbased on the large number of inputs. The back propagationalgorithm is a probabilistic hill climbing algorithm that isutilized to train the neural networks. In the neural networks theneurons process the information using a connection basedapproach for computation. The back propagation is utilized forsending the features in the forward direction while propagatingthe errors in the backward directions so that the minimum errorsamples are selected. However the use of back propagationalgorithm which uses weight in the network to determine theerror functions, suffers from the step-size problem as itcomputes only the partial first derivative. Back propagationgenerally is considered to take infinitesimal steps to reduce theerror at each step and achieves a minimum error function. Butpractically infinitesimal steps are not taken and only the largerstep sizes are chosen which do not converge to a good solution.This problem can be resolved by utilizing the slope of errorfunctions and curvature of the higher-order derivatives inweight space. But unfortunately back propagation algorithmdoes not include that information.

In order to generate dataset with minimum error samples usingminimum error function, the neural network is trained byextreme learning machine (ELM) instead of the backpropagation algorithm. The ELM approach randomly assignshidden node parameters, the input weights and biases foradditive hidden nodes. Then the hidden layer output matrix iscalculated and the output weight vector is obtained usingwhich the feature dataset can be generated by solving theregularized least mean square optimization. However thegenerated feature subset has redundancy and hence a moresparse solution is required which can be achieved byintroducing hierarchical ELM (H-ELM) [15].

H-ELM is built in a multi-layer manner unlike the greedylayer-wise training of traditional frameworks. H-ELM istraining architecture is divided into two separate phases:unsupervised hierarchical feature representation and supervisedfeature classification. ELM-based auto-encoder is developedfor extracting the multi-layer sparse features of the input datawhile the ELM-based regression is utilized for the finaldecision making in the generation of the dataset. The optimallyselected features and the statistical measured features whichare fused together are transformed into an ELM random featurespace in order to exploit the hidden information among thetraining samples. Then aN-layer unsupervised learning isperformed to obtain the high-level sparse features withminimum errors. The output of each hidden layer isrepresented as

Hi=g(Hi-1.ß) → (7)



Where Hi is the output of the ith layer, Hi-1 is the output of the(i-1)th layer, g(.) denotes the activation function of the hiddenlayers and β is the output weights.

In H-ELM, each hidden layer is an independent module andthe functions are a separated feature extractor. When the layersincreases the resulting features become compact. Once thefeature of the previous hidden layer is extracted, the parametersof the current hidden layer will be fixed, and need not be fine-tuned. As retraining the NN is iteratively performed, thetraining using H-ELM is faster than traditional methods.

After the unsupervised hierarchical training is completed, theresultant outputs of the kth layer are considered as the high-level features extracted from the input feature space. Theseextracted features are randomly perturbed and are utilized asinput to the supervised ELM-based regression to obtain thefinal feature sets that can be included in the dataset. H-ELM isdeveloped on the random feature mapping for enhancing thelearning speed and exploits the universal approximationcapability of the ELM in both feature learning andclassification. Thus H-ELM reduces the number of neuralnodes and improves the testing time while also effective inretaining the lost information for the best generation of datasetwith minimum error samples.

Association rule mining using frequent pattern-growth (FP-growth) algorithmAssociation rules are the probabilistic relationship between twosets of data or features. In mobile based diabetes diagnosis, theassociation rules are generated to determine the relationbetween the training and testing feature sets. Apriori algorithmhas been previously utilized for the detection of associationsbetween the feature sets using the association rules generated.Each set of data has a number of items and is called atransaction. The output of Apriori is the sets of rules thatdetermine the frequency of items are contained in sets of data.However, for the candidate generation the apriori algorithmutilizes more memory and time and also costly especially whenthere are large number of patterns. Hence in the mobile baseddiabetes diagnosis system, the apriori algorithm is replaced bythe FP-growth algorithm [16]. FP-growth algorithm does notgenerate the candidate sets like apriori and instead employs apattern-fragment growth for determining the association rules.

In FP-growth algorithm, the larger feature space is condensedinto smaller sub-spaces so that the costly repeated scans areavoided. A set of features (items) are taken and the databasegenerated by using NN is considered. The support of a patternis determined and then the minimum support threshold ξ ispredefined using which the frequent patterns are detected andthe associations between the features are exploited.

Algorithm 3: Frequent-pattern Growth algorithm

Input: Dataset D generated by H-ELM based NN; minimumsupport threshold ξ

Output: Complete set of frequent patterns

Call FP-growth (FP-tree, null)

Let P be the single prefix-path part of Tree

Q be the multipath part with the top branching node replacedby a null root

Freq_pattern_set(P) and Freq_pattern_set(Q) be the generatedset of patterns

FP-growth (Tree, α)

If Tree contains single prefix path then

For each combination β of the nodes in the path P do

Generate pattern ß U α with support = minimum support ofnodes in β

Freq_pattern_set(P)

End for

Else assign Q as Tree

For each items (features) k in Q do

Generate pattern ß=k U α with support = k.support

Construct β’s conditional pattern-base and then β’s conditionalFP-tree Treeβ

If Treeβ = null then

Call FP-growth (Treeβ, β)

Freq_pattern_set(Q)

End if

End for

End if

Return(Freq_pattern_set P ∪ Freq_pattern_set Q ∪ Freq_pattern_set P × Freq_pattern_set QThe three frequent pattern sets generated using the FP-growthalgorithm, are considered to be the association rules thatdetermine the relation between the feature sets. Using therelations detected between the features, the presence orabsence of diabetes can be diagnosed effectively and the resultsare displayed in the mobile phones.

Experimental ResultsIn this section, the performance of the proposed Mobile-baseddiabetes diagnosis system is evaluated in terms of accuracy,precision, recall and time consumed. The evaluation results arecompared with methods namely Diabetes retinopathy [5] andDiagnosis using BP-NN& apriori algorithm [17] to determinethe efficiency of the proposed diagnosis system. The input dataare taken from real-time dataset collected from patients andfrom STARE database along with the data extracted from

Sridar/Shanthi


glucometer results. The better diagnosis method must help indetecting the diabetes accurately and in less time. Theexperimental results are conducted and comparison graphs aregiven below in Table 1.

Table 1. Performance evaluation results.

Methods Accuracy (%) Precision Recall Time (min)

Diabetesretinopathy

84 0.80 0.78 32

Diagnosis usingBP-NN & apriorialgorithm

89 0.86 0.85 18

Mobile-baseddiabetes diagnosissystem

96 0.91 0.93 12

AccuracyAccuracy is defined as the exact diagnosis of the diseasewithout any false detection.

Figure 2. Accuracy comparison.

Figure 2 shows the comparison of the diagnosis methods interms of accuracy. When the data are processed, the accuracyof diabetes retinopathy method is 84%, Diagnosis using BP-NN & apriori algorithm is 89% and the proposed Mobile-baseddiabetes diagnosis system is 96%. Hence it is clear that theMobile-based diabetes diagnosis system outperforms othermethods in terms of accuracy during diagnosis of diabetes.

PrecisionPrecision is the measurement of refinement or exactness of thediagnosis system.

Figure 3 shows the comparison of the diagnosis methods interms of precision. When the data are processed, the precisionof Diabetes retinopathy method is 0.8, Diagnosis using BP-NN& apriori algorithm is 0.86 and the proposed Mobile-baseddiabetes diagnosis system is 0.91. Hence it is clear that theMobile-based diabetes diagnosis system outperforms othermethods in terms of precision during diagnosis of diabetes.

Figure 3. Precision comparison.

RecallRecall is the relevancy in the output of the diagnosis systemwith respect to the presence of diabetes.

Figure 4. Recall comparison.

Figure 4 shows the comparison of the diagnosis methods interms of recall. When the data are processed, the recall ofdiabetes retinopathy method is 0.78, Diagnosis using BP-NN &apriori algorithm is 0.85 and the proposed Mobile-baseddiabetes diagnosis system is 0.93. Hence it is clear that theMobile-based diabetes diagnosis system outperforms othermethods in terms of recall during diagnosis of diabetes.

Time consumptionIt is the time taken for a diagnosis system to diagnose thepresence of diabetes from the given input data. Figure 5 showsthe comparison of the diagnosis methods in terms of timeconsumed. When the data are processed, the diabetesretinopathy method takes 32 minutes, Diagnosis using BP-NN& apriori algorithm takes 18 minutes and the proposed Mobile-based diabetes diagnosis system takes 12 minutes. Hence it isclear that the Mobile-based diabetes diagnosis systemoutperforms other methods and processes data within less time.Thus from the experimental results it can be concluded that theproposed Mobile-based diabetes diagnosis system performsbetter than the other compared methods in terms of accuracy,precision, recall and time consumption for diagnosis.



Figure 5. Time consumption.

ConclusionMobile-based diabetes diagnosis system proposed in this paperhelps in diagnosing the pre-diabetes and also in detecting thecurrent status of the patients from the home itself in a moreconvenient manner. The proposed Mobile-based diabetesdiagnosis system utilizes data from glucometer, retinal imagesof patients and the clinical data gathered to detect the type Iand Type II diabetes effectively. The Features are extractedfrom the input data and the optimal features are selected usinga modified PSO. Then the statistical measures are calculatedand these features are merged with the optimal features. Thefeature space thus formed is processed in H-ELM based NN togenerate a dataset with minimum error samples. Then theassociation rules are generated using FP-growth algorithm todetermine the relation between the features In order to detectthe presence of diabetes in the specified patient data.Experimental results also show that the proposed Mobile-baseddiabetes diagnosis system helps in diagnosing diabetes moreaccurately.

References1. Bourouis A, Feham M, Hossain MA, Zhang L. An

intelligent mobile based decision support system for retinaldisease diagnosis. Decision Support Systems 2014; 59:341-350.

2. Karan O, Bayraktar C, Gümüşkaya H, Karlık B.Diagnosing diabetes using neural networks on small mobiledevices. Expert Syst Appl 2012; 39: 54-60.

3. Nes AA, van Dulmen S, Eide E, Finset A, KristjánsdóttirÓB, Steen IS, Eide H. The development and feasibility of aweb-based intervention with diaries and situationalfeedback via smartphone to support self-management inpatients with diabetes type 2. Diabetes Res Clin Practice2012; 97: 385-393.

4. Erkaymaz O, Ozer M. Impact of small-world networktopology on the conventional artificial neural network forthe diagnosis of diabetes. Chaos Solitons Fractals 2016; 83:178-185.

5. Roychowdhury S, Koozekanani D, Parhi K. Dream:Diabetic retinopathy analysis using machine learning. IEE JBiomed Health Informat 2014; 18: 1717-1728.

6. Torkaman M, Bagheri G, Ahmadi A, Saburi A. AComparative Study of Blood Glucose Measurements UsingGlucometer Readings and the Standard Method in theDiagnosis of Neonatal Hypoglycemia. Iran J Neonatol2016; 7: 41-46.

7. Gaikwad S, Jadhav V, Gurav A, Shete AR, Dearda HM.Screening for diabetes mellitus using gingival crevicularblood with the help of a self-monitoring device. J PeriodontImplant Sci 2013; 43: 37-40.

8. Bin-Sabbar MS, Al-Rodhaan MA. Diabetes monitoringsystem using mobile computing technologies. Diabetes2013.

9. Singhal S, Ralhan P, Jatana N. Smartphone-basedcolorimetric detection to measure Blood Glucose Levels. InContemporary Computing (IC3), Eighth InternationalConference, IEEE, 2015.

10. Chikh MA, Saidi M, Settouti N. Diagnosis of diabetesdiseases using an artificial immune recognition system2(AIRS2) with fuzzy k-nearest neighbor. J Med Syst 2012;36: 2721-2729.

11. Zhang B, Kumar BV, Zhang D. Detecting diabetes mellitusand nonproliferative diabetic retinopathy using tonguecolor, texture, and geometry features. IEEE TransactBiomed Eng 2014; 61: 491-501.

12. Silva S, Martins H, Valente A, Soares S. A bluetoothapproach to diabetes sensing on ambient assisted livingsystems. Procedia Comput Sci 2012; 14: 181-188.

13. Zhu W, Hou D, Zhang J, Zhang J. Optimization of a subsetof apple features based on modified particle swarmalgorithm. In Intelligent Information Technology andSecurity Informatics (IITSI), Third InternationalSymposium, IEEE, 2010.

14. Tseng YJ, Ping XO, Liang JD, Yang PM, Huang GT, Lai F.Multiple-Time-Series Clinical Data Processing forClassification With Merging Algorithm and StatisticalMeasures. IEEE J Biomed Health Informa 2015; 19:1036-1043.

15. Tang J, Deng C, Huang GB. Extreme learning machine formultilayer Perceptron. IEEE Transact Neural NetworksLearning Syst 2016.

16. Han J, Pei J, Yin Y, Mao R. Mining frequent patternswithout candidate generation: A frequent-pattern treeapproach. Data Mining Knowledge Discov 2004; 8: 53-87.

17. Sridar K, Shanthi D. Medical Diagnosis System for theDiabetes Mellitus by using Back Propagation-AprioriAlgorithms. J Theor Appl Informa Technol 2014.

*Correspondence toK Sridar

Department of Computer Science and Engineering

Odaiyappa College of Engineering and Technology

India

Sridar/Shanthi


Documents

Design and development of mobile phone based … · Design and development of mobile phone based diabetes mellitus diagnosis system by using ANN - FP-growth techniques. K.Sridar1*,