CLASSIFICATION OF SUBJECTS WITH CHOLESTEROL USING IRIS

International Research Journal of Engineering and Technology (IRJET) e ISSN: 2395 0056

Volume: 09 Issue: 05 | May 2022 www.irjet.net p ISSN: 2395 0072

CLASSIFICATION OF SUBJECTS WITH CHOLESTEROL USING IRIS

1PG Student, Department of Electronic and Instrumentation, Bharathiar University, Coimbatore, Tamil Nadu, India

2 Associate Professor and Head, Department of Electronic and Instrumentation, Bharathiar University, Coimbatore, Tamil Nadu, India

3 Ph.D Research Scholar, Department of Electronics and Instrumentation, Bharathiar University, Coimbatore, Tamil Nadu, India ***

Abstract - Cholesterol, natural substance present inhuman body, a fatty component required for most of the metabolic activity. Level of cholesterol greater than 200 mg/dL is considered as high cholesterol level and leads tocriticalillness such as hypertension, coronary heart disease etc. Iridologyis a type of ComplementaryandAlternative Medicine, non invasive diagnostic tool having the ability to detect the cholesterol at early stage. The concept of iridology is the analysis features of iris in determining the defective organ. This work combined the concept of iridology and deep learning, promising method in medical analysis, in identifying subjects with high cholesterol. In this study deep learning classifiers such as AlexNet, VGG 16, ResNet 18 and ResNet 50 were used for classifying subjects with high cholesterol and healthy.

Key Words: AlexNet, Cholesterol, Iridology, ResNet 18, ResNet 50,VGG 16

1. INTRODUCTION

Cholesterol is a fatty component present in the body. Cholesterol plays a vital role in carrying out various metabolic activities such as production of vitamin D and testosterone,helpsincarryingoutdigestionprocessetc.[1]. Cholesterolcirculatesinbloodusinglipoproteinpresentin the blood. High density lipoprotein (HDL), Low density lipoprotein(LDL)andVeryLow DensityLipoproteins(VLDL) aretypesoflipoproteinspresentinhuman.HDLandLDLare often termed as good and bad cholesterol. If the concentration of cholesterol increases in blood it leads to varioushealthissuessuchascoronaryheartdisease,stroke, peripheralarterialdiseaseetc.Almost75%ofcholesterolin blood stream is produced by liver and remaining are obtainedfromthefood.Levelbetween125to200mg/dLis consideredtobeanidealcholesterollevelinbloodvalue.The risk factor increases when the cholesterol level increases above 200mg/dL and this is termed as high cholesterol condition.Thehighcholesterolconditionwasdiagnosedby analyzingthebloodfromsubjectsafterundergoing10to12 hoursoffasting[2].

Iridologyisconsideredasanalternativediagnostictoolfor detecting the defect organ in human. In the concept of iridology, the iris was divided over 60 regions each representingaspecificorgan.Itanalysisthefeaturesofiris

based on the shape and structure of iris to determine the defectiveorgan[3].Highcholesterolcanbeidentifiedbythe presenceofcholesterolringorsodiumringontheiriswhich isofgreyishorwhitishthatisvisibleontheedgesoftheiris oftencalledasArcusSenilis.Itiscausedbythefatdepositin theinnerliningoftheperipheralcornea[4]. Fig 1showsthe sampleimageofarcussenilisorsodiumring.

Deeplearningisasubsetofmachinelearningtechniquethat mimicshumanactivity.Inthistechniqueacomputermodel learnstoperformclassificationtasksdirectlyfromimages, text, or sound. Models are trained using a large set of labelleddataandneuralnetworkarchitecturesthatcontain manylayers.Theterm“deep”usuallyreferstothenumberof hidden layers in the neural network. Traditional neural networks only contain 2 3 hidden layers, while deep network can have as many as 150 layers [5]. This study combined the concept of iridology and deep learning to classifysubjectswithhighandnormalcholesterollevelusing irisimage.

Fig-1:ImageofArcusSenilis[6]

2. RELATED WORK

R.A.Ramlee et al. [7] proposed Otsu’s threshold based cholesterol detection system. Circular Hough Transform (CHT), parabolic Hough transform and Daugman's Rubber SheetModelfordetectionofiris,eyelidsandnormalizationof iris,respectively.Imagesofirisforhighcholesterollevelwere collected from Mediscan and National Library of Medicine websites. Iris images from CASIA, MMU, UPOL and UBIRIS datasetswereconsideredashealthy.

DianSarietal.[9]developedaFuzzyLocalBinaryPattern basedfeatureextractionsystemfordetectingcholesterol.The featureswereobtainedfromcroppedirisimagefordetecting cholesterol. The iris images were captured using mobile

2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal

International Research Journal of Engineering and Technology (IRJET) e ISSN: 2395 0056

camera from a hospital to carry out this study. 105 iris imageswereobtainedfromsubjectsofthese63and42iris imageswereusedtotrainandtestthenetwork.Thesystem obtainedaccuracy,precision,recallandcomputingtimeas 95.23%90.47%,100%,and40ms,respectively.

Amini N et al. [10] in their work verified that AlexNet obtained100%accuracyindeterminingirisimageswithand withoutsodiumring.Ridza AzriRamleeetal [11].Intheir studyfoundthatthe30%irisimageswithmorethan139as threshold values is considered as the distinguish factor

betweenthehealthyandunhealthyirisimagesaftertesting 30healthyimages.VikasBhangdiyaetal.[12]foundthatas theeigenvalueincreasesthecholesterolvaluealsoincreases.

Sruthi K et al. [13] in their study developed a 33 layer Convolutional Neural Network (CNN) to classify images amonghealthy,riskofcholesterolandhighcholesterollevel. The CNN based classification system obtained accuracy of 100%,99.21%and100%forhealthy,riskofcholesteroland highcholesterollevel,respectively.

Fig 3

:

Processinvolvedinclassificationofhealthyandunhealthyirisimage

Thesignificancesoftheproposedworkareasfollows:

(i)Employedpre trainedmodelsuchasAlexNet,VGG 16, ResNet 18 and ResNet 50 for classifying iris images betweenhealthandunhealthy.

(ii)Comparedtheemployedexistingpre trainedmodel basedonaccuracyinclassifyingirisimagesintoheathyand unhealthybasedonthesodiumring.

3. METHODOLOGY

This section explains the work flow of the proposed methodology.Fig 2showsthepre processingstepsinvolved intheproposedalgorithm.

InputIris Image Image Resize RGBTO GRAY

Fig 2:Pre processingtechniqueinvolvedinproposed Work

3.1 Pre-Processing

3.1.1 Input Iris Image

The human iris images used in the proposed work are obtainedfromIITD(IITDelhiIrisDatabase),severalwebsites andprivatehealthcenter.TheIITDimagewasconsideredas healthycategoryandwasusedtotrainandtestthenetwork. About50arcussenilisimageswereobtainedfromvarious onlinemedicalwebsitesand149imageswereobtainedfrom a private clinic using Logitech webcam. Of the 149 images 102 belong to unhealthy category and 47 to the healthy category.

3.1.2 Image Resize

Theimageswereobtainedfromvarioussourcessotheywere ofdifferentsizes. Alltrainingandtestingimagesshouldbeof samesizetotrainandtestthedeeplearningarchitecture.So, alltheusedimageswereconvertedto227X227forAlexNet and224X224forVGG 16,ResNet 18andResNet 50.

Volume: 09 Issue: 05 | May 2022 www.irjet.net p ISSN: 2395 0072 © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal |

International Research Journal of Engineering and Technology (IRJET) e ISSN: 2395 0056

Volume: 09 Issue: 05 | May 2022 www.irjet.net p ISSN: 2395 0072

3.1.3 RGB TO GRAY

The resized image is converted into a grey scale image to decrease the complexity of the process and as the image obtainedfromIITDdatasetareNIRimages.Allthegrayscale convertedimageshavethechannelas3.

3.2 Proposed Technique

This subsection provides the detailed architecture of the networkusedforclassificationofcholesterol.Fig 3shows the process involved in classification of healthy and unhealthyirisimagesbasedonthepresenceofsodiumring.

All the pre processed images were separated into healthy and unhealthy iris images. Five fold cross validation technique is employed to train and test the pre trained networks. In cross validation technique each fold will be consideredasatestsetexactlyi.e.,whenone foldactastest set while the remaining will act as training set. The set whichyieldswithmaximumaccuracywillbeconsideredto testwhichareobtainedfurther.

In the proposed work AlexNet, VGG 16, ResNet 18 and ResNet 50pre trainedarchitecturewereused.Totrainand testAlexNettheimageswereconvertedto227 X227X 3. The VGG 16, ResNet 18 and ResNet 50 were trained and testedwithimagesofresolution224X224X3.TheAlexNet consistsofeightlayersofthesefiveareconvolutionallayers andthreearefullyconnectedlayers. VGG 16consistsof16 layers.FirsttwolayersinVGG 16consistof64channelswith filtersizeof3X3.Thethirdlayerisamax poollayerwith strideas(2,2)withconvolutionallayerof256channeland filtersizeof(3,3)followedbyamax poollayerassameas previouslayer. Twoconvolutionallayeronewith3X3as filtersizeand256aschannelandamax poollayerfollowed by3X3filtersizewith512channelandsamepadding.The images are then passed to cascade of 2 fully connected layers. ResNet 18 consists of eighteen layers of deep CNN with72layers.ResNet 50is a typeofResNetarchitecture havingforty eightconvolutionlayerswithonemax pooland an average pool layer and consists 3.8×109 floating point operations.

The data are simulated in the computer with Intel core i5 10500hCPU@4.5GHZ6cores,64 bitWindows11operating system,16GBRAMDDR44000MHZwithGPUofNvidiaRTX 3060 6GB GDDR6, SSD of 512GB equipped with MATLAB R2021aenvironment.Theepochisgivenas100,batchsize as 10 and learning rate as 0.0001 for all the pre trained models.

4 RESULTS AND DISCUSSION

Theoriginalimageoftheirisisfirstresized(227X227and 224X224)byusingtheimagebatchprocessingapp,thenthe resized images are converted from RGB image to the grayscaleimage,thus pre processingisdone.Theimageis

then trained by each of the network (Alexnet, VGG 16, Resnet 18andResnet 50),thebelowimage(Fig 4)shows theresultofthepre processedstep(originalimage,resized image,andthegrayscaleconvertedimage).Fig 5andFig 6 showstheclassificationresultobtainedforthissystem.

Fig 4:Pre Processedimage

Fig 5:Finalresultsoftheunhealthyimage

Fig 6:Finalresultsofthehealthyimage

Table 1 shows the accuracy of the proposed network in classifyingtheimagebetweentwoclasses,normalimageand thehighcholesterolimage.Thesystemaccuratelypredicts normal and high cholesterol. The reduction of accuracy in VGG 16 and the ResNet 18 is due the epoch assigned. The accuracycanbeincreasedbyincreasingthepre setofthein builtfunction‘Maxepoch’.

International Research Journal of Engineering and Technology (IRJET) e ISSN: 2395 0056

Volume: 09 Issue: 05 | May 2022 www.irjet.net p ISSN: 2395 0072

Table 1:ClassificationofAccuracy

NETWORK ACCURACY

ALEXNET 100%

VGG 16 99.4% RESNET 18 99.84% RESNET 50 100%

Theproposedclassifytheimageintohealthyandcholesterol basedonthepresenceofsodiumring.Inthiswork,theimage batchprocessingwasusedtoresizeandtoconverttheRGB imagetogray.TheAlexnet,VGG 16,ResNet 18,andRestNet 50 was used for classification. An accuracy of 100% was obtainedforAlexnetandResnet 50.Accuracyof99.4%and 99.84%wasobtainedforVGG 16andResNet 18,respectively in categorizing the iris image into normal and high cholesterol. The performance of the proposed system was comparedwithotherexistingsystemsandwastabulatedin Table 2.Thisworkisanelementaryworkextensivedatabase has to becreatedtoprove the effectiveness of iridologyin detectingcholesterol.

Table 2:ComparisonofAccuracyoftheproposedsystem withtheexistingsystem

Sl. No. Authors Classes Classified Accuracy (%)

1 MelvinDanieletal.[8] 3 94.6

2 DianSarietal.[9] 3 95.23

3 AminiNetal.[10](Alex) 2 100

4 AminiNetal.[10](CNN) 2 98

5 AminiNetal.[10](VGG 16) 2 98.5

6 SruthiKetal.[13] 3 99.21

7 Proposed system (Alexnet, ResNet 50) 2 100

7 Proposed system (ResNet 18) 2 99.84

7 Proposed system (VGG 16) 2 99.4

5 CONCLUSIONS

Thisworkcomparedtheefficiencyofthepre trainedmodel in categorizing the iris images into healthy and high cholesterol based on accuracy. This work found out that whentheepochvalueisincreasedtheclassificationaccuracy also increases. This system obtained 100% accuracy for AlexNetandResNet 50.ForResNet 18andVGG 16obtained the accuracy of 99.84 and 99.4, respectively. Accuracy of

thesenetworkscanbeincreasedbyincreasingtheepoch.The performance of the employed pre trained models was comparedamongthemandwithotherpre existingsystem from which we inferred that AlexNet and ResNet 50 has betterefficiencycomparedtoothermodels.Thiswork isa preliminaryworkalargenumberofdatahastobecollected toprovetheefficacyofiridologyindetectingcholesterol

6 ACKNOWLEDGEMENTS

TheauthorswouldalsoliketoacknowledgeDr.R.Aishwarya and all subjects who helped develop the database. The authorsalsoliketothankIITDforprovidingtheirisdatabase

REFERENCES

[1] Cholesterol https://medlineplus.gov/

[2] R. Enjelica, “Deteksi Kelebihan Kolesterol melalui Citra Iris Mata dengan Metoda Discrete Wavelet Transform dan Klasifikasi K Nearest Neighbor,” UniversitasTelkom,Bandung,2019.

[3] Iridology https://www.clinicaladvisor.com/home/features/a lternative meds update/iridology detecting impaired organ function with the iris/

[4] Andana, Shafira Nur, Ledya Novamizanti, and IN Apraz Ramatryana. "Measurement of Cholesterol ConditionsofEyeImageusingFuzzyLocalBinary Pattern(FLBP)andLinearRegression."2019IEEE International Conference on Signals and Systems (ICSigSys).IEEE,2019.

[5] Deep Learning https://in.mathworks.com/discovery/deep learning.html.

[6] https://commons.wikimedia.org/wiki/File:Arcu_Se nilis.jpg

[7] Ramlee, R. A., et al. "Automated detecting arcus senilis,symptomforcholesterolpresenceusingiris recognition algorithm."Journal of Telecommunication, Electronic and Computer Engineering(JTEC)3.2(2011):29 39.

[8] Daniel,Melvin, JangkungRaharjo,and Koredianto Usman."Iris basedimageprocessingforcholesterol level detection using gray level co occurrence matrix and support vector machine."Engineering Journal24.5(2020):135 144.

[9] Sari, Dian, Jangkung Raharjo, and Ledya Novamizanti."CholesterolLevelDetectionThrough Eye Image Using Fractal and Decision Tree."IOP Conference Series: Materials Science and Engineering.Vol.982.No.1.IOPPublishing,2020.

2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal

International Research Journal of Engineering and Technology (IRJET) e ISSN: 2395 0056 Volume: 09 Issue: 05 | May 2022 www.irjet.net p ISSN: 2395 0072

[10] Amini,N.,andA.Ameri."Adeeplearningapproach toautomaticrecognitionofarcussenilis."Journalof biomedicalphysics&engineering10.4(2020):507.

[11] Ramlee, Ridza Azri, et al. "Detecting cholesterol presencewithirisrecognitionalgorithm."Biometric Systems,DesignandApplications(2011):129 149

[12] Bhangdiya,Vikas."Cholesterolpresencedetection usingirisrecognition."Int.J.Technol.Sci1.1(2014): 22 25

[13] Sruthi,K.,J.Vijayakumar,andS.Thavamani."Deep Learning based Automatic Detection of Cholesterol."2021 Smart Technologies, CommunicationandRobotics(STCR).IEEE,2021.

[14] Ramlee,R.A.,andS.Ranjit."Usingirisrecognition algorithm, detecting cholesterol presence."2009 International Conference on Information ManagementandEngineering.IEEE,2009.

[15] Amini,N.,andA.Ameri."Adeeplearningapproach toautomaticrecognitionofarcussenilis."Journalof biomedicalphysics&engineering10.4(2020):507.

[16] Shanker, N. R., A. Ezhil, and S. Archana. "Non invasive method of detection of cholesterol using imageprocessing."InternationalJournalofMedical EngineeringandInformatics4.3(2012):223 230.

2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal