Retinal Image Lesions Assisted Diabetic Retinopathy Screening System Through Machine Learning

Abstract

This paper proposes an Automatic Screening System that identifies the symptoms of Diabetic Retinopathy from retinal fundus images and determines the severity. The entire system is of four-fold; the first phase pre-processes the retinal images and eliminates uneven illuminations. The second phase segments the major parts of images such as optic Disk and Blood Vasculature. Once the segmentation is completed, the residual part is called as lesion image and it is subjected to a hierarchical classification in the third phase. It is a two-step classification method which classifies the entire image into four regions namely Hard Exudates, Microaneursyms, Hemorrhages, cotton wool spots. For classification, three classifiers such as k-Nearest Neighbor, Gaussian Mixture Model and Support vector Machine are analyzed. Finally, the classified regions are processed through an objective function for severity grading. The novelty of this paper is a composite feature extraction method which extracts different and diverse features from lesion image. Since the lesions are very small and of different shapes, the proposed feature extraction method can provide more discriminant information for classifier. For experimental validation, a standard and publicly available DIARETDB1 dataset and the performance is measured through sensitivity, specificity, accuracy and area under curve. On an average, the proposed method gained accuracy, sensitivity, and specificity of 90.2 %, 95.2 % and 82.1 % respectively.

Multi-label classification of retinal disease via a novel vision transformer model

“…In general, the study of Berbar ( 2022 ) can grade the severity of DR into three different levels. Recently, the study of Reddy and Ravindran ( 2022 ) presented an automatic screening platform to recognize DR in retinal images. The proposed classification scheme consists of two phases.…”