Deep Learning Based Diagnostic-Aid System For Ophthalmic (Eye) Diseases

Abstract

Diabetic Retinopathy is a global health problem, affecting more than 100 million people globally. It is the most prevalent complication caused by diabetes and has always been one of the leading causes of blindness. Early detection holds significant importance in safeguarding patients from vision loss and irreversible retinal damage by slowing the progression of the disease and treating it in time. With early diagnosis and treatment, the risk of severe vision loss is reduced by 90%. However, early detection remains a challenging task, as patients are often asymptomatic until proliferate stages, and many lesions are often undetectable to the human eye in their early stages and therefore easily overlooked, even by a retinal specialist. In this work, we propose a system that automates the diagnosis of Diabetic Retinopathy using deep learning techniques, particularly convolutional neural networks (CNNs), and aims to identify its presence in its early stages. Additionally, a retinal blood vessel segmentation system using the U-Net architecture was developed to enhance the analysis of the vascular layer of the eye by ophthalmologists and surgeons, aiding in the treatment of Diabetic Retinopathy and other diseases affecting the vascular layer. The detection model achieved up to 93.1% in confidence of correctly grading the stage of Diabetic Retinopathy.