Deep Learning Approach for Classification of Retinal Diseases from Optical Coherence Tomography Images

Wang, Pei Chee (2023) Deep Learning Approach for Classification of Retinal Diseases from Optical Coherence Tomography Images. Final Year Project (Bachelor), Tunku Abdul Rahman University of Management and Technology.

Text Full Text - WANG PEI CHEE.pdf Restricted to Registered users only Download (8MB)

Abstract

Image classification of retinal diseases is crucial and plays an important role when it comes to diagnosing early stages of retinal diseases to prevent the worsening of retinal diseases that might lead to irreversible vision loss. However, excessive domain knowledge and experience are required to accurately diagnose retinal diseases due to the features of each retinal diseases are difficult to be differentiated and recognized by human eyes and misinterpretations might happen. Hence, a CNN system that is able to assist experts is needed. Therefore, the implementation of existing CNN models is trained, validated, and tested using a publicly obtained dataset from Kaggle. There are three main objectives of this project. Firstly, to implement a CNN model for 4 classes of retinal diseases, which are the Diabetic macular edema (DME), Choroidal Neovascularization (CNV), Drusen, and Normal. Secondly, to evaluate the proposed CNN model with existing model in terms of model accuracy. Thirdly, to develop a web application that allows ophthalmologists to refer diagnostic reference. Transfer Learning of EfficientNet-B3 exhibited the highest model accuracy of 99.7% and 0.3% of misclassification. Hence, it is selected to be the proposed solution for this project.