Advancing Optical Tomography Image Analysis: Exploring Convolutional Neural Network Model Variants for Retinal Damage Detection

Retina is a vital component of the eye; it plays a crucial role in visual perception. Any damage to the retina can lead to vision impairment and severe ocular diseases. Early detection and accurate diagnosis of retinal conditions is crucial. In the domain of medical image analysis, especially within the scope of retinal imaging, convolutional neural network (CNN)-based deep learning models have exhibited promising results. This paper involves the detection of retinal conditions such as Diabetic Macular Edema (DME), Choroidal Neovascularization (CNV) and Drusen. The paper presents a comparative study of CNN models for the detection of these retinal damages using Optical Coherence Tomography (OCT) images. This study focuses on evaluating the performance of popular CNN architectures, like Residual Networks (ResNet), Dense Convolutional Networks (DenseNet), Inception Networks, Visual Geometry Group (VGG). Also, the paper provides a comparative analysis of different depth variants of deep learning models such as ResNet (18, 50 and 101), DenseNet (121, 169, and 201), Inception-V3, Inception-ResNet-V2, and VGG (16 and 19). These models are assessed on a dataset containing OCT images of Normal, DME, CNV and Drusen retinal conditions. The effectiveness of the models are evaluated using evaluation metrics such as accuracy, precision, recall and F1 score. Additionally, visualizations such as confusion matrix, training accuracy versus validation accuracy plots, training loss versus validation loss plots are utilized to provide a better understanding of the model performance. The idea is to explore the impact of depth on the model performance and understand how deeper models demonstrate an enhanced ability to learn complex features and capture intricate details within the retinal data. Exploring these depth variants aims to provide insights into the advantages and potential trade-offs associated with increasing depth of deep learning models for retinal damage detection tasks.