Utilizing Transfer Learning for Automated Detection of Diabetic Retinopathy

Abstract

Diabetic retinopathy (DR) occurs when diabetes mellitus damages the blood vessels in the retina which is the part of the eye responsible for capturing images. This damage can cause vision problems or even lead to blindness if not addressed in time. Early detection and treatment are key to prevent serious vision loss from DR. However, diagnosing DR through manual examination of fundus images is both costly and time-consuming due to the need for trained medical professionals. Therefore, the development of automated DR detection systems is needed. This paper explores the use of transfer learning models, specifically VGG16, ResNet18, and AlexNet, on Fourier-Bessel series expansion (FBSE) based dyadic decomposition (FBD) decomposed images to detect DR. Transfer learning is crucial in medical image analysis since, it enables the development of accurate and efficient models even with limited data and computational resources. Network parameters are fine tuned and better features are extracted by applying the proposed approach. The results are computed over the benchmark dataset of DR. It is observed that our proposed approach based on ResNet18 network is giving the best performance and 92.15% accuracy is achieved. © 2024 IEEE.