VisionAiNet: An Architecture for Advancing Ocular Disease Diagnosis Utilizing CNN

Ocular diseases represent a significant public health concern worldwide, which needs accurate and efficient diagnostic methods for timely intervention and management. Fundus imaging, a foundation of ophthalmic diagnostics, offers a non-invasive yet comprehensive view of the retina, insights into disease severity and early warnings, aiding in the identification and characterization of various ocular pathologies. However, the manual classification of these images by ophthalmologists may require substantial time investment and is subject to inter-observer variability. The advancement of artificial intelligence (AI) and machine learning algorithms has revolutionized medical imaging analysis, presenting an opportunity to enhance the diagnostic process. Deep learning, renowned for its excellence in tasks related to computer vision, forms the foundation of our strategy. In this study, we analyzed the efficiency of our newly developed architecture VisionAiNet in detecting ocular diseases, namely cataract, diabetic retinopathy, glaucoma, and normal eye conditions. The study reveals that while pretrained models offer competitive performance, a scratch CNN (VisionAiNet) architecture consistently outperforms in disease detection with the accuracy of 94.83%, highlighting the importance of tailored model development, contributing to the advancement of automated diagnostic systems in ophthalmology.