Deep Learning-Based Classification for Eye Diseases using OCT Images
Roni Davidov MSc thesis poster
Abstract
Optical Coherence Tomography (OCT) is widely used for diagnosing and monitoring retinal diseases and structural biomarkers such as Choroidal Neovascularization (CNV), Diabetic Macular Edema (DME), and drusen. However, manual interpretation of OCT scans is time-
consuming, subjective, and sensitive to inter-observer variability, particularly in large clinical datasets and in cases involving subtle pathological changes.
This study presents a deep learning–based framework for automated classification of retinal diseases and biomarkers from OCT images. The approach progresses from single-slice (2D) analysis using CNN models to patient-level multi-slice (3D) modelling. A hierarchical classification strategy is introduced to separate initial screening from fine-grained diagnosis, reflecting clinical decision pathways and improving sensitivity in screening tasks. Multi-slice information is incorporated using hybrid CNN architectures combined with BiLSTM or Transformer-based aggregation. Experimental results demonstrate that transfer learning improves training stability, while multi-slice modelling enhances patient-level performance.
Grad-CAM visualizations indicate that the model focuses on clinically relevant retinal structures.