Analysis of Retinopathy Images to Detect the effect of Diabetes on Eye

Abstract

Diabetic Retinopathy (DR) is one of the most occurred diabetic eye diseases that cause serious complication to the eye even to vision loss. The risk of this problem can be completely prevented by means of two fundamental public health interventions such as early diagnosis and treatment. However, the effective screening and diagnosing the DR from the retinal fundus images is a major challenge. In this scenario, the experts and other professional diagnosis teams depend on analyzing the retinal fundus images using the Computer-Aided Diagnosis (CAD) system. The general processing stages of CAD system involves pre processing, segmentation and classification pertaining to retinal blood vessels or other abnormalities. Various machine learning methods have been developed. Further the recent contribution of deep learning models and its successful performance over the conventional techniques under medical applications have been motivated the researchers for adopting the deep learning models to diagnose the DR. This research work focuses to analyze the retinal fundus images for the effective diagnosis of DR using intelligent techniques. It performs the DR detection in different phases. In the first phase the blood vessel analysis is performed by, image enhancement, filtering and morphological operation. In the filtering, Peak Signal-to Noise Ratio (PSNR) and Mean Squared Error (MSE) computation is performed on for DRIVE database. Blood vessels are segmented using morphology operation. Feature extraction by DWT and Gray-Level Co-Occurrence Matrix (GLCM) is done and classification by Support Vector Machine (SVM) and distance measures like city block, Spearman, and Minkowski are employed for CHASE_DBI and DRIVE databases. In the second phase, the examination of optic disc, exudates and blood vessels is done. Here, wavelet transform is used for detecting the optic disc using DIARETDB1 dataset, whereas morphological operation, grey level thresholding and Discrete Wavelet Transform (DWT) is used for detecting the exudates and classify the image into normal and abnormal. The third phase of the research work segments and analyses the blood vessels for detecting DR. Initially, the combination of Contrast Limited Adaptive Histogram Equalization (CLAHE) and average filter performs the pre-processing. Further, the optimized Gray level thresholding is used for the blood vessel segmentation and feature extraction techniques like Texture energy Measurement (TEM), Local Binary Pattern (LBP), Shanon’s entropy and Kapur’s entropy are utilized. The classification of the image is performed by the optimal vii trained Neural Network (NN). The developmed new algorithm termed as Modified Levy Updated-Dragonfly Algorithm (MLU-DA) enhances the performance of both segmentation and classification to classify the image as normal and abnormal using the High Resolution Fundus (HRF) images. The fourth phase of this research work diagnoses the DR from the retinal fundus images by segmenting and analyzing the retinal abnormalities like haemorrhages, Microaneurysm, soft exudates and hard exudates. After enhancing the contrast of the image by CLAHE, open-close watershed transformation helps to remove the optic disc and Grey Level thresholding helps to remove the blood vessels. Further, the utilization of Top hat transformation followed by Gabor filtering segments the retinal abnormalities. From the segmented abnormalities, LBP, TEM, Shanon’s, and Kapur’s entropy are extracted as features and carried out the optimal feature selection by proposed Modified Gear and Steering-based Rider Optimization Algorithm (MGS-ROA). Further, the optimized Deep Belief Network (DBN) classifies the image into normal, severe, moderate, and earlier stages of DR, in which MGS-ROA updates the DBN weight. The experiment is carried out for DIARETDB1image datasets. The final contribution of this research work focuses on analyzing the optic disc, blood vessels and retinal abnormalities for diagnosing DR using DIARETDB1 image datasets The open-close watershed transform segments the optic disc, Grey level thresholding segments the blood vessels and top hat transform followed by Gabor filtering segments the abnormalities like haemorrhages, Microaneurysm, soft exudates and hard exudates. Here also, features like LBP, TEM, Shanon’s, and Kapur’s entropy are extracted from the segmented optic disc, blood vessels, and abnormalities and the proposed Trial-based Bypass Improved Dragonfly Algorithm (TB-DA) is used to carry out the optimal feature selection process The classification of images is done by the hybridization of DBN and NN, in which the TB-DA optimizes weight function of both classifies