A Hybrid Deep Learning Approach With Explainable AI for Diabetic Retinopathy Classification

This project develops a hybrid deep learning framework for diabetic retinopathy (DR) classification from retinal fundus images, comparing four architectures: Swin Transformer, CoAtNet (Convolutional Transformer Network), a custom CNN-Transformer Hybrid combining EfficientNet-B0 with a two-layer transformer encoder (86%), and DeiT (Data-efficient Image Transformer). To ensure clinical interpretability, three complementary explainable AI techniques—Grad-CAM (Gradient-weighted Class Activation Mapping), LIME (Local Interpretable Model-agnostic Explanations), and Integrated Gradients—are applied to visualize and localize DR-related features such as exudates, hemorrhages, and microaneurysms. This approach enables transparent, trustworthy automated screening suitable for clinical decision support.