Smartsight: Android-Based Retinal Disease Detection using Retinal Fundus Images with XAI

Diabetic retinopathy (DR) is a serious eye condition affecting diabetic patients, stemming from microvascular damage to retinal blood vessels. If not diagnosed and treated early, DR can lead to blindness. Traditional diagnostic methods involve manual examination of retinal images by ophthalmologists, which is labor-intensive and time-consuming. This highlights the need for automated, efficient, and accurate diagnostic tools. This study proposes a deep learning-based framework using ResNet50 architecture enhanced with Gradient-weighted Class Activation Mapping (Grad-CAM) for the automatic detection and classification of diabetic retinopathy from retinal images obtained via a smartphone-based fundus camera. The ResNet50 model, known for its deep residual learning capabilities, was trained on a dataset of retinal images to differentiate between healthy and DR-affected eyes. Incorporating Grad-CAM provides visual explanations by highlighting the retinal image regions most indicative of DR, thereby improving model interpretability. The proposed model demonstrated high performance, achieving 98% accuracy on the training dataset, 91% on the validation dataset, and 96% on the test dataset. These results highlight the model's robustness and potential for facilitating early DR detection, especially in resource-constrained environments with limited access to specialized ophthalmic care.