ENHANCING DIABETIC RETINOPATHY DIAGNOSIS WITH GRAPH NEURAL NETWORKS
Keywords:
Diabetic retinopathy, graph neural networks, variational auto encoders, retinal image classification.Abstract
Leading reason for blindness internationally, diabetic retinopathy (DR) requires exact conclusion to help brief
medicines. Clinicians' manual fundus imaging assessment is work escalated and inclined to botches. Automating
DR finding utilizing PC helped advancements — particularly Convolutional Neural Networks (CNNs) — show
guarantee. This paper presents a Graph Convolutional Neural Network (GCNN) technique to further develop
retinal picture handling, to be specific focusing on infection seriousness characterization. GCNNs increment
highlight extraction by utilizing topological connections inside pictures, accordingly creating more exact
arrangement results. Accuracy, precision, recall, and F1-score among assessment estimates show the
recommended GCNN model's effectiveness. With an accuracy of 89% on the doled out dataset, trial results
show the GCNN model beats current strategies. Additionally, the review investigates a few Transfer Learning
(TL) models including InceptionV3 and Xception, hence creating accuracy paces of 92%. This study gives
specialists a reliable and viable strategy for computerized finding, subsequently supporting early DR
distinguishing proof and mediation. The venture likewise recommends making an easy to use front-end interact
with the Flask framework and including security client testing through validation.
