Detection and Classification of Diabetic Retinopathy Using Modified Inception V3

Abstract

In the last decade, the prevalence of diabetic retinopathy (DR), a sight-threatening medical condition of diabetes mellitus, has markedly increased, impacting millions globally. The conventional method of diagnosing and classifying this condition was done through physical and detailed examination of fundus images by ophthalmologists, a process prone to human error and time-consuming. To overcome this challenge, artificial intelligence, particularly deep learning algorithms, has taken a position in automating the diagnosis of diabetic eye disease and categorization from fundus images. Various studies have confirmed the effectiveness of convolutional neural networks in this task, with Inception V3 emerging as a particularly successful architecture. In our current work, we adduce a novel approach for DR detection and categorizing it utilizing the Inception V3 architecture on fundus images. The learning rate is modified between le-3, le-4, le-5, and le-6 to investigate various optimization approaches. Our model, trained on the Asia Pacific Tele Ophthalmology Society datasets, accomplished an accuracy of 91.64% for a learning rate of le-5, which outperforms current approaches in diagnosing the five phases of DR.

Keywords

• deep learning
• convolutional neural networks
• retina images
• inception v3
• diabetic retinopathy