A LIGHTWEIGHT DEEP LEARNING SYSTEM FOR IDENTIFYING DIFFERENT TYPES OF EYE DISEASES USING EXPLAINABLE ARTIFICIAL INTELLIGENCE

Early and accurate detection of ocular diseases is essential to prevent irreversible vision loss, particularly in regions with limited access to ophthalmic specialists. Deep learning has demonstrated remarkable success in medical image analysis; however, the high computational requirements and limited interpretability of many state of the art models restrict their deployment in resource-constrained settings. This paper presents a lightweight and explainable deep learning framework for classifying five common ocular conditions cataract,glaucoma, uveitis, strabismus, and normal eyes using images captured with consumer-grade devices. The proposed system employs MobileNetV2 with a two-phase fine tuning strategy,class balanced training,and extensive augmentation to enhance generalization. Furthermore, Gradient-weighted Class Activation Mapping(Grad CAM)is integrated to generate visual explanations,improving model transparency and clinical trust. Experimental results on the Ocular Disease Recognition dataset demonstrate an accuracy of 93% while maintaining minimal computational cost,highlighting the suitability of the system for mobile and low-power platforms. The findings suggest that lightweight CNN architectures, combined with explainability,can serve as practical diagnostic support tools in underserved healthcare environments.