Abstract:
Given the worldwide prevalence of eye diseases, we need to develop new ways for diagnosing them, particularly in resource-poor areas. Recent deep learning models for ophthalmology generally have a lower coverage of diseases, do not provide explainable predictions and are not patient centered. This study presents FusionEyeNet, a comprehensive web-platform in addressing these limitations with LLM-powered explaining external eye disease classification and personalized recommendation. The proposed system uses a hybrid architecture that combines MobileNetV2 and VGG16 by using strategic feature fusion, which makes it feasible to classify five external eye conditions efficiently such as Cataract, Conjunctivitis, Eyelid disorders, Normal eyes and Uveitis. Explainable AI is incorporated into the framework using Grad-CAM, providing a clear diagnostic reasoning with its focus on medically related areas in ocular images. Furthermore, Google Gemini large language model also provides personalized educational suggestions according to diagnostic results and confidence scores. The experimental results confirm the competitive superiority of FusionEyeNet with 97.89% test accuracy, outperforming the comprised (individual) baseline models such as VGG16 by a clear margin (97.19%). The system realizes accurate detection of cataract and robust diagnosis to the minority class, such as uveitis. The implemented web-app allows users to upload eye images and get diagnosed including confidence scores, as well as visual explanations with AI-generated educational advice. These findings create a benchmark for reliable ophthalmic AI systems, which integrate both diagnostic performance and educational value, with the potential to improve eye care accessibility and guide clinical management worldwide.
