Deep Learning Models for Ocular Disease Detection

Abstract

Early and precise identification of ocular diseases using fundus imaging is essential for effective ophthalmic care, yet it remains a non-trivial challenge due to the complexity of visual patterns and inter-observer variability. Conventional manual diagnosis, while widely practiced, is labor-intensive and susceptible to subjective bias, thereby limiting scalability and consistency in clinical settings. To address these limitations, automated diagnostic systems based on deep learning have emerged as a promising alternative, offering improved efficiency, reproducibility, and diagnostic accuracy. In this work, we propose a robust deep learning framework for automated detection of ocular diseases, with a specific focus on distinguishing between normal and cataract cases. This was performed through two deep learning models, an InceptionV3 structure and a pre-trained GoogLeNet model, which was developed and extensively validated. Both models were trained and evaluated on the Ocular Disease Intelligent Recognition (ODIR-2019) dataset in case of binary classification tasks. The InceptionV3 model had a relatively high accuracy of 97.9\% and the GoogLeNet model also achieved a high accuracy of 97.6\%. These results demonstrate the possibility of professional DL algorithms to provide comprehensible, precise, and efficient solutions for automation diagnosing ocular diseases, and to be a significantly innovative breakthrough to ophthalmic disease detection and clinical decision supporting.