International Conference on Intelligent Sustainable Systems · Mar 4, 2026
A Deep Learning-Driven Framework for Multi-Ocular Disease Classification and Instance Segmentation in Retinal Imaging
By Panugan FM, Refulle LE, Villamil PJ et al.
The Accurate and efficient analysis of retinal fundus images is critical for the early detection and management of ocular diseases. This study proposes a deep learning-driven framework for the simultaneous classification and instance segmentation of multiple ocular pathologies using YOLOv12-seg. The model was trained and evaluated on a curated retinal fundus dataset comprising age-related macular degeneration, pathologic myopia, and cataract-related blur artifacts, with custom pixel-level annotations generated to support instance segmentation. An$\mathbf{8 0}-\mathbf{1 0}-\mathbf{1 0}$train-validation-test split was employed to ensure robust evaluation. Experimental results demonstrate strong performance, achieving an mAP@0.5 of 0.93 and an mAP@0.5:0.95 of 0.767, along with balanced precision, recall, and mean Intersection over Union scores, indicating reliable localization and accurate segmentation of pathological regions. Comparative analysis against established architectures, including U-Net and Mask R-CNN, shows that the proposed framework delivers superior accuracy while maintaining near real-time inference through its single-stage design. Qualitative visualizations and deployment-level testing further confirm stable performance under diverse imaging conditions. These findings highlight the effectiveness of YOLOv12-seg as a practical and scalable solution for automated multi-ocular disease screening, supporting its potential integration into AI-assisted ophthalmic diagnostic workflows.