Hybrid Lion-Cat Optimized Deep Learning Model for Retinal Image-Based Diabetic Retinopathy Detection

Abstract

The significance of Diabetic Retinopathy (DR) detection on time is absolutely crucial as it aids in identifying the vision threatening disease that deteriorates the blood vessels of retina in a wider range. The present study involves in detecting the DR through the implementation of image processing approach that is capable enough in identifying the disease with maximal accuracy and reliability. The retinal images are pre-processed in the initial stage by eliminating the unnecessary noises with the aid of Hilbert Transform (HT). The noise free images are then effectively segmented into multiple regions with the aid of Artificial Neural Network (ANN) based approach for effectively assessing the stages of disease. Then, the Speeded-Up Robust Features (SURF) is preferred in this work for extracting the features of the segmented images in an efficient manner. Eventually, the optimal features are selected and classified by implementing a hybrid Lion-Cat optimized Convolutional Neural Network (CNN). The performance measures of this CNN with hybrid optimization approach is analogized to the like Support Vector Machine (SVM), ANN and conventional CNN, which validates that the proposed approach delivers optimal accuracy than the other prevailing approaches in identifying the DR on time.