CNN-Based System for Diabetic Retinopathy Detection

Bhawalkar, Sayali Vishwas (2025) CNN-Based System for Diabetic Retinopathy Detection. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (959kB) | Preview
Preview	PDF (Configuration Manual) Download (352kB) | Preview

Abstract

Diabetic retinopathy (DR) is the cause of sight loss among the diabetic patients across the world. Appropriate classification and early detection of DR severity are vital at ensuring that the condition is treated early and intervention is performed. This project aims at suggesting a diagnostic pipeline that lies the efficacy of deep learning algorithms to deliver EfficientNetsB0 convolutional neural network in order to classify the retinal fundus pictures into one of the five severity phases of DR. The system utilizes a pretrained EfficientNetB0 model trained in APTOS 2019 resulting in high classification accuracy with a macro F1-score of 68.63 % and accuracy of 83.90. Grad-CAM visualizations were incorporated into the models to help make them more interpretable, visualizing areas of the retina that impact the prediction and provide more clinical optionality when using the system. Additionally, the entire solution was also deployed as an interactive web application wherein users can upload images (in real-time) and generate class predictions as well as Grad-CAM overlays. Common challenges of deep learning methods include class imbalance, overfitting when using smaller datasets, overall high-cost computationally, and low interpretability; this paper attempts to overcome these problems through the use of data augmentation and transfer learning, choosing EfficientNetB0 because of its efficiency, and using Grad-CAM because of how it provides transparency. The proposed work shows that high-performance deep learning models can be combined with explainable AI and human-friendly deployment to assist with DR screening both in the clinical and remote facilities.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Garg, Mohit UNSPECIFIED
Subjects:	R Medicine > RE Ophthalmology Q Science > QH Natural history > QH301 Biology > Methods of research. Technique. Experimental biology > Data processing. Bioinformatics > Artificial intelligence Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Artificial intelligence R Medicine > Healthcare Industry
Divisions:	School of Computing > Master of Science in Artificial Intelligence
Depositing User:	Ciara O'Brien
Date Deposited:	11 Aug 2026 15:36
Last Modified:	11 Aug 2026 15:36
URI:	https://norma.ncirl.ie/id/eprint/9499

Actions (login required)

View Item