Ocular Disease Recognition System using ResNet50 and InceptionV3 over DenseNet, Xception, VGG and U-Net Architectures

Abstract

Millions of people worldwide continue to face pre- ventable visual deficit problems arising from undiagnosed eye diseases like diabetic retinopathy, glaucoma, and cataracts. The situation on the ground is even more abysmal in rural areas. The project introduces an Al-based Ocular Detection System, which acts as a sophisticated machine to bridge this gap in healthcare by streamlining the detection of common ocular diseases from retinal fundus images. Using deep learning neural networks in addition to ensemble methods, the system looks forward to increasing the diagnostic accuracy by cutting down the diagnosis time in stress-challenged areas while ensuring improved access to care for the most underserved populations. Automating the image analysis from the retina's perspective by improving the very foundation of rural healthcare enables early diagnosis and, hence, improve vision health outcomes for the patients. The system makes use of the effectiveness of two deep learning models, InceptionV3 and ResNet50, which are well-known for their strong feature extraction capabilities and high accuracy in medical image analysis. These models maintain computational efficiency appropriate for real-time deployment while enabling accurate multi-disease detection from retinal fundus images.