Diabetic Retinopathy Detection Using SVM

Abstract

Innovation is getting progressed step by step in pretty much every field. This work includes the detection of Diabetic Retinopathy (DR). Diabetes happens when the pancreas neglects to emit sufficient insulin, and gradually influences the retina of the natural eye. As it advances, the vision of a patient begins deteriorating (depleting), prompting diabetic retinopathy. In such manner, retinal pictures gained through fundal camera help in investigating the outcomes, nature, and status of the impact of diabetes on the eye. The main aim of this study is Age-related Macular Degeneration (AMD) through Local Binary Patterns (LBP) and further trial and error utilizing Gray-Level Co-Occurrence Matrix (GLCM). For this reason, the presentation of Gray level Co-Occurrence Matrix (GLCM) as a surface descriptor for retinal pictures has been investigated and contrasted and different descriptors, for example, GLCM filtering (GLCMF) and local phase quantization (LPQ). This will take to the of blood vessel highlights, for example, energy, differentiation, correlation and homogeneity values. We involve SVM as a classifier to recognize valid and bogus vessels. The conclusion of diabetic retinopathy depends on clinical eye assessment and eye fundus imaging.

Introduction

I. INTRODUCTION

Diabetic retinopathy causes visual deficiency. Diabetic retinopathy is one of the eye illnesses which is caused because of retinal vein extraction. Diabetic retinopathy influences veins in the light-touchy tissue that is known as the retina that lines to the rear of the eye. It is the most normal reason for vision misfortune among individuals with diabetes and the main source of vision hindrance and visual deficiency among working-age grown-ups. Here we proposed a framework where we extricated retinal veins for distinguishing eye infections.  Physically extricating the retinal veins is a monotonous undertaking and there are many mechanized strategies are accessible which makes work more straightforward. Here we proposed a calculation to separate veins from retinal pictures.  We utilized sifting systems to dispose of commotion and ecological impedance from picture. Neighborhood entropy thresholding and different progressive Channel procedure has been taken on during this framework. We had executed this framework in MATLAB. Client will include retina picture into framework. Framework will apply sifting strategies. Picture preprocessing steps are applied to come by exact outcome. Framework will eliminate all undesirable articles from picture. Framework will apply this calculation to extricate retinal blood vessels. At long last, framework will identify diabetic retinopathy.

II. LITERATURE SURVEY

In the paper presented by Nikita Gurudath, Mehmet Celenk,and H.Bryan Riley, they've developed an approach to automate the identification of the presence of diabetic retinopathy from color fundus images of the retina has been proposed. They achieved an bracket of an input fundus image into one of the three classes, healthy/ normal, on-Proliferative Diabetic Retinopathy (NPDR) and Proliferative Diabetic Retinopathy (PDR) also blood vessel segmentation from the input image was achieved by them by use of Gaussian filtering. A versatile, input - driven approach is considered for the mask generation and thresholding was satisfied utilizing original entropy.

Their exploration incorporates a three concentrated artificial neural network (ANN) and support vector machines (SVM) to classify the retinal images [1].In the paper presented by Kranthi Kumar, Palavalasa and Bhavani Sambaturu, a system has been introduced which uses a new system to descry hard exudates with high delicacy with respect to lesion level.They originally detected the possible seeker exudate lesions by using the aft ground deduction methodology in the present system. Following the posterior way, in the last stage of the algorithm they removed the false exudate lesion findings using thede-correlation stretch grounded system. Likewise they tested algorithm on intimately available DiaretDB database, which contains the ground verity for all images [2].

In the paper presented by Niladri Sekhar Dattaa, Himadri Sekhar Dutta, Mallika De, and Saurajeet Mondal, they proposed a system using position grounded discrepancy improvement process which is popularly known as Differ Limited Adaptive Histogram Equalization (CLAHE) for the discovery of retinal changes in DR images.

Their proposed algorithm divides the retinal images into a number of small  on-overlapping contextual penstocks. Likewise, following CLAHE at each pipe independently, median filtering of DR images is carried out in order to smooth the background noise. Results of their proposed algorithm showed a considerable enhancement in the improvement of DR image quality. The outgrowth generated by them for average perceptivity and particularity of this Diabetes Webbing System revealed as 82.64 and 99.98 independently [3].In the paper presented by Usman M. Akram, Shoab A. Khan, they proposed a computer aided system for the early detection of DR. They presented algorithms for retinal image preprocessing, blood vessel enhancement, segmentation, optic disk localization and detection which eventually lead to detection of different DR lesions using proposed hybrid fuzzy classifier.They tested developed methods on four different publicly available databases. After which they compared presented methods with recently published methods and the results show that presented methods outperform all others[4].

In the paper presented by ShitalN. Firke and Ranjan Bala Jain, it presents a complication neural network approach to detecting diabetic retinopathy. The use of intimately accessible Apatos Blindness Discovery database to train a complication neural network was done, where the image is reused at an early stage, primarily involving image resizing, pixel rescaling, and marker encoder. After that, an image is given to the complication neural network model, to decide whether the case has diabetic retinopathy or not. About 3789 color retinal images are used in trials to train the proposed model and about 948 images are collected to test its effectiveness in bracket [5].In the paper presented by Anam Tariq,M. Usman Akram, Arslan Shaukat and ShoabA. Khan, they developed a system that excerpts the macula from digital retinal image using the vascular structure and optical slice position. It creates a double chart for possible exudate regions using sludge banks and formulates a detailed point vector for all regions. The system uses a Gaussian Mixture Model- grounded classifier to the retinal image in different stages of maculopathy by using the macula coordinates and exudate point set. The results attained from their system have been compared with other styles in the literature in terms of perceptivity, particularity, positive prophetic value and delicacy [6].

In the paper presented by Gehad Hassan, Nashwa El-Bendary, Aboul Ella Hassanien, Ali Fahmy, AbullahM.Shoeb and Vaclav Snasel, they presented blood vessel segmentation approach, which can be used in computer grounded retinal image analysis to prize the retinal image vessels. They used the Mathematical morphology and K- means clustering to member the vessels. To enhance the blood vessels and suppress the background information, they perform a smoothing operation on the retinal [7].In the paper presented by Jayant Yadav, Manish Sharma and Vikas Saxena, they approached the idea issue by using computer vision to not only descry this complaint, but also automating this procedure using a neural network to give results of numerous cases within a short time frame. DR discovery is being done by detecting two major corridor of it, videlicet, Exudates and FleckHemorrhages.They designed the overall armature of the design in following ways 1. Rooting Optic Disk 2. Rooting Blood Vessels 3. Discovery of Exudates 4. Discovery of Fleck hemorrhages [8].

In the paper presented by Karkhanis Apurva Anant, Tushar Ghorpade and Vimla Jethani, the proposed fashion uses image processing, data mining, texture and sea features which are uprooted for discovery. The results are attained for image of standard database DIARETDB1 and estimated using parameters of perceptivity, particularity and delicacy. The final outgrowth generated by them yields a high delicacy of 97.75 which can help to descry and help diabetes [9].In the paper presented by Xianglong Zeng, Haiquan Chen, Yuan Luo and Wenbin Ye,a computer- backed opinion system grounded on deep literacy algorithms is proposed to automatedly diagnose the referable diabetic retinopathy (RDR) by classifying color retinal fundus photos into two grades. In this work, a new convolutional neural network model with Siamese-suchlike armature is trained with transfer literacy fashion. The proposed model accepts binocular fundus images as inputs and learns their correlation to help make prognostications. To further corroborate the effectiveness of the binocular design, a binocular model for five- class DR discovery is also trained and estimated on a 10 confirmation set [10].

III. PROPOSED SYSTEM

Fig 1. shows the detail system architecture. It gives the details description for every method and algorithms we have used in this work. Diabetic retinopathy is a GUI which shows the detail results for the given input image. This approach will be very useful for diabetic patients by detecting diabetes at early stage. For this work we will be majorly using MATLAB and Supervised learning algorithms. The entire project is divided into two major parts first is image pre processing for feature extraction and second is machine learning part which uses pattern recognition. Firstly we need to provide an input image that image is fed into the system. After loading image first step is to be carried out is RGB conversion, input image is converted into three different images of red, green and blue. Further green image is taken for processing where filters are applied and image is re sized to a particular format and noise is removed. Next step is to convert the image into histogram equalized image by converting it to gray scale image. From this image features such as blood vessel extraction, exudates and Microaneurysms are extracted. These features are stored into database. Now the main part is to select the required feature for further classification. In this process accuracy measures are tested and dataset is trained. We use supervised learning method as dataset is unlabeled, for that we uses K-means and fuzzy C-Means(FCM) for classification purpose. And lastly validation process is done to check whether patient suffers from from diabetes or not by using support vector clustering algorithm. Finally result obtained is stored into multi class SVM.

A. Modules

1. RGB Conversion 
2. Image Pre-processing
3. Image Enhancement
4. Blood Vessel Extraction
5. GLCM Feature Extraction

IV. RESULT AND DISCUSSION
This framework is assessed on a huge dataset containing 100 retinal pictures. It brings about getting the elements of veins. Our algorithm has the advantage that it is  applicable to all types of retinal images

This is a simple and interactive GUI we have created in Matlab. Firstly by clicking Load button you can load the image. We have developed a window using x-axis and y-axis where the selected image will be loaded. We have used circle finder method for optimizing the exact input image within the given range of co-ordinates. By clicking process button Operation will be performed and output plots are displayed. And then according to the given plots accuracy parameters are given which gives the assurance for given result

As fig. 5 shows the result for given input image. We can see that after processing and classifying the image we have obtained the result as “Patient is Healthy”.  Hence no plots are displayed. It simply means the patient doesn't suffer from diabetes.

Fig.6 shows the another scenario where result is different. Here after processing the given image we have obtained the result that “Patient suffers from diabetes”. After this result plots will be displayed which include Green Channel image, Histogram Equalized image, Binary image, Optic disk, Microaneurysms and Hemorrhages. SVM constructs an optimal hyper plane that will maximize the margin of separation between the classes. The feature vector that lie close to the margin are called the support vectors.

The Fig.7 shows the first image we have obtained through RGB Conversion. After converting an input image into RGB we get three different images of red blue and green. From these images we will select green channel image for further classification. The other two images that is red and blue are more saturated and less contrast Green channel image is more optimized and more suitable for feature extraction. Also the given image goes through different databases the size of an images may vary so we need to obtain a particular size for input image. For this purpose we use a filter called median filter. It helps to remove hue and noise from the image. Median filter is best known order statistic filter which replaces value of pixels by the median of intensity values in the neighborhood of that pixel. This green channel image is further converted into gray scale image.

The given gray scale image is then converted into histogram equalized image by implementing image enhancement techniques. The main idea behind this is to bring out the hidden details of the image. Histogram equalized image gives you the clear vision through which detection of Microaneurysms, optic disk and hard exudates is possible. For this purpose we use CLAHE algorithm which is a variant of adaptive histogram equalization. It is used for over amplification of the contrast. It works on small regions of eye rather than entire image, these regions are called as tiles. The side by tiles are then combined by using bi-linear interpolation to remove

Fig 9. shows the binary image which is converted from histogram image. The image is transformed into black and white series of pixels. The aim of this step is to highlight the other features such as optic disc detection, exudates detection and Microaneurysms. It also gives the view of blood vessels which are affected by hemorrhages. Segmentation of the image is done here. It is done by thresholding and it divides the image into objects or regions. This divides image into foreground and background. Detection of optic disk is required because it has same intensity and brightness with other features i.e. exudates in retinal image. Here region is thresholded and optic disk is being detected by proper boundary.

Dilation is very important factor in retinopathy. Dilation causes bright regions to be highlighted and dark regions are ignored. Dilation of morphological operation is used for detection of exudate edges. It is performed on green channel image of two different size. Also subtraction is done between two sizes to get the edge detection of the exudates.

The term Red Lesion is also known as Microaneurysms and hemorrhages. There can be different size of lesions which needs to be focused also false detection on blood vessels needs to be performed. For this purpose filters are used to differentiate between red lesions and blood vessels.

Conclusion

An unique method for AMD and DR diagnosis was presented. It is based on an- analyzing texture discrimination capability ties in fundus images to differentiate healthy patients from AMD and DR images. In future, these diseases will experience a high growth due to diabetes in- crease. The performance of GLCM along with different classifiers was tested and compared with other texture descriptors. The most important finding is that the proposed method is capable of discriminating the classes based on analyzing the texture of the retina back- ground, avoiding previous segmentation of retinal lesions. This can be a great re- search for future.

References

[1] Nikita Gurudath, Mehmet Celenk, and H.Bryan Riley, “Machine Learning Identification of Diabetic Retinopathy from Fundus Images”, Signal Processing in Medicine and Biology Symposium (SPMB),IEEE-2014 [2] Kranthi Kumar Palavalasa and Bhavani Sambaturu “Automatic Diabetic Retinopathy Detection Using Digital Image Processing “,International Conference on Communication and Signal Processing(ICCSP), IEEE-2018 [3] Niladri Sekhar Datta,Himadri Sekhar Dutta , Mallika De and Saurajeet Mondal, “An Effective Approach: Image Quality Enhancement for Microaneurysms Detection of Non-Dilated Retinal Fundus Image “, International Conference on Computational Intelligence: Modeling Techniques and Applications (CIMTA),ElSEVIER-2013 [4] Usman M. Akram and Shoab A. Khan, “Automated Detection of Dark and Bright Lesions in Retinal Images for Early Detection of Diabetic Retinopathy”, Springer-2012 [5] Shital N. Firke and Ranjan Bala Jain “Convolutional Neural Network for Diabetic Retinopathy Detection”, International Conference on Artificial Intelligence and Smart Systems (ICAIS),IEEE-2021 [6] Anam Tariq, M. Usman Akram, Arslan Shaukat and Shoab A. Khan “ Automated Detection and Grading of Diabetic Maculopathy in Digital Retinal Images ”, Journal of Digital Imaging,Springer-2013 [7] Gehad Hassan, Nashwa El-Bendary, Aboul Ella Hassanien, Ali Fahmy, Abullah M.Shoeb and Vaclav Snasel “Retinal blood vessel segmentation approach based on mathematical morphology”,International Conference on Communication, Management and Information Technology (ICCMIT), ElSEVIER-2015 [8] Jayant Yadav, Manish Sharma and Vikas Saxena “ Diabetic Retinopathy Discovery using feedforward Neural Network”, Tenth International Conference on Contemporary Computing (IC3), IEEE-2018 [9] Karkhanis Apurva Anant, Tushar Ghorpade and Vimla Jethani “Diabetic Retinopathy Detection through Image Mining for Type 2 Diabetes “ , International Conference on Computer Communication and Informatics (ICCCI),IEEE-2017 [10] XIANGLONG ZENG, HAIQUAN CHEN, YUAN LUO and WENBIN YE “Automated Diabetic Retinopathy Detection Based on Binocular Siamese-like Convolutional Neural Network”, IEEE Access,2019 [11] M. Abirami, M. Vignesh, K. Vikramsriram and E. Shivanithyesh “Diabetic Retinopathy Detection using DeepLearning Techniques”,International Research Journal of Engineering and Technology (IRJET),2021 [12] Sagar Suresh Karki and Pradnya Kulkarni, “A Systematic Review on Diabetic Retinopathy Detection Methods”,International Journal of Advance Research in Computer Science and Management Studies,IJARCSMS-2015 [13] Ahmed Wasif Reza, C. Eswaran and Kaharudin Dimyati “Diagnosis of Diabetic Retinopathy: Automatic Extraction of Optic Disc and Exudates from Retinal Images using Marker-controlled Watershed Transformation”, Journal of Medical Systems,Springer-2010 [14] Sarika B. Patil and B.P. Patil “ Retinal fundus image improvement using adaptive CLAHE styles”, journal of Seybold Report, [15] R. Naveen, S. A. Sivakumar, B. Maruthi S)ankar and A. Keerthana Priyaa, “Diabetic Retinopathy Detection using Image Processing” ,International Journal of Engineering & Advanced Technology (IJEAT)

Copyright

Copyright © 2022 Mrs. Vishakha Shelke, Mr. Vinay Manish Shah, Mr. Harsh Ratnani, Mr. Rahul Despande. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.