Author Country (or Countries)



In this paper, a novel method is developed to establish a framework for Investigation on Retinal Fundus Images for Detection of Diabetic Retinopathy and Classification. The preprocessing is done through green-channel enhancement followed by top-hat filtering method to enhance image details for subsequent segmentation and feature extraction. The extracted features are used to train the database images using Artificial Neuro Fuzzy Interference System. The optic disc and the blood vessels are found using a supervised segmentation algorithm, damaged area and hard and soft exudates using Kirsch operator to extract the features for the classification of healthy and abnormal images of Diabetic Retinopathy from the retinal images as Proliferal Diabetic Retinopathy (PDR) or Non-proliferal Diabetic Retinopathy (NPDR). The NPDR is further classified into mild, moderate and severe cases based on the calculation of microaneurysms count using Local Thresholding (LT), Local Shifted Thresholding (LST) and the count is compared to the Global Thresholding (GT) to provide the best classification results. Results are optimized, in terms of their sensitivity, specificity, accuracy and $Q$ factors by calculating the True Positive (TP), True Negative (TN), False Positive (FP) and False Negative (FN) analysis of the test image. The images are trained with Artificial Neural Network (ANN) and Artificial Neuro Fuzzy Interference System (ANFIS). Analyzed results are compared and validation set is obtained for both methods.

Digital Object Identifier (DOI)