Alam, Md. AshrafulHossain, MD. TamzidBhowmik, UtsavMila, Riza AsmatChowdhury, Mahtab ShahriarKarmakar, Ronak2025-01-142025-01-14©20242024-01ID 19101121ID 19101646ID 20101590ID 19101637ID 18101298http://hdl.handle.net/10361/25150Cataloged from PDF version of thesis.Includes bibliographical references (pages 40-42).This thesis is submitted in partial fulfillment of the requirements for the degree of Bachelor of Science in Computer Science, 2024.Diabetic retinopathy is one complicated eye complication of diabetes and considered one of the major causes of preventable blindness worldwide. Diabetic retinopathy occurs when high glucose levels in the blood damage small blood vessels of the retina over time continuously, resulting in various problems with vision. In its early stages, DR typically shows no symptoms; thus, early detection is very important in order to avoid permanent loss of vision. Given the importance of early diagnosis, advanced machine learning systems, especially those applying deep learning, have been very important in eye care in recent times. This work presents a new deep learning model using ensemble learning combined with a hybrid architecture and proposes a deep learning model named DRDetector. The proposed DRDetector combines ResNet50 for feature extraction with Vision Transformer ViT layers to understand the global context. This methodology overcomes the challenge of diagnosis and prediction of diabetic retinopathy with enhanced accuracy while minimizing false positive and negative cases. DRDetector uses a Convolutional Neural Network (CNN) combined with a Vision Transformer architecture, with transfer learning for detection of DR stages. It classifies the retinal images into different classes including healthy, and different stages of DR: mild, moderate, NPDR, and PDR. The aim of this paper is to comprehensively assess the performance of DRDetector based on a large dataset of retinal images, so that its efficacy can be shown in clinics. This would lead to improved diagnosis with higher accuracy, reduction of diagnostic errors, and in effect, help the ophthalmologists39; quest for perfection. Moreover, an advanced grading system can assist healthcare practitioners in grading the severity of the disease for better management and treatment options for DR. This study has pointed out that optimized deep learning systems may support early detection, risk evaluation, and personalized treatment for diabetic retinopathy patients.50 pagesenBrac University theses are protected by copyright. They may be viewed from this source for any purpose, but reproduction or distribution in any format is prohibited without written permission.Diabetic retinopathyDeep learningEye complicationDisease detectionPredictive analysisDiabetic retinopathy--Diagnosis.Neural networks (Computer science).Machine learning.Diabetes--Complications--Diagnosis.Artificial intelligence--Medical applications.Thesis