Comparative Analysis for Diabetic Retinopathy Detection Using CNN and Transfer Learning
Abstract
Diabetic Retinopathy is an eye disease that is seen in patients suffering from diabetes. The manual diagnosis process for diabetic retinopathy is using Funduscopy by some healthcare professional or with the physician analyzing the patient's retina on the monitor. Automating this manual procedure will act as a second opinion to the health professionals, can be used in rural areas where efficient healthcare facilities are not available. Many types of research on automatic systems like Computer Aided Diagnostic (CAD) System are currently in trend. In this article, we have discussed two strategies for achieving the Automatic Diagnostic System for Diabetic Retinopathy detection from fundus images: Convolutional Neural networks (CNN) and Efficient Net. CNN proves to be an efficient method for low illuminance images.EfficientNet-B5 architecture is inputted with pre-processed images for the classification step. The experimentation shows that EfficientNet-B5 gives better results than CNN models.