Publication:
AI powered early detection of diabetic retinopathy using fundus images

Loading...
Thumbnail Image
Date
2024-08
Authors
Norsyamilah Azahar
Journal Title
Journal ISSN
Volume Title
Publisher
Research Projects
Organizational Units
Journal Issue
Abstract
This research addresses the urgent need for early detection of diabetic retinopathy (DR) using Artificial Intelligence (AI) and fundus image analysis. DR, a leading cause of blindness in diabetic patients, highlights the criticality of timely diagnosis and treatment. AI, particularly Convolutional Neural Networks (CNNs) like ResNet-34, advances this effort by accurately categorizing retinal images into Normal Retina, Mild, Moderate, Severe Non-Proliferative DR, and Proliferative DR categories. These systems leverage deep learning to detect subtle retinal abnormalities, enabling prompt intervention to halt disease progression. AI's scalability allows efficient analysis of large datasets, improving diagnostic accuracy and clinical decision-making. Comparative analysis with the referenced paper “ResNet-34/DR” reveals insights into model performance. In this study, ResNet-34 achieved 93.08% training accuracy and 81.42% validation accuracy, showing gradual improvement across training epochs despite observed discrepancies likely due to dataset variations. In contrast, the referenced paper reported higher and more balanced accuracies around 95.00%, indicating superior generalization and reduced overfitting. Both studies noted misclassifications, especially in Moderate and Proliferative DR, yet the referenced paper demonstrated fewer errors and higher average metrics (accuracy, precision, recall, and F1-score), highlighting its robust classification capability. These findings underscore opportunities to optimize the model in this study for enhanced performance metrics, thereby improving its clinical utility in diagnosing and managing DR.
Description
Keywords
Citation