Panoptic-Net: A Unified Deep Learning System for Classifying the Full Spectrum of Retinal Disease from Fundus Photographs

Background/Objectives: Vision impairment due to retinal diseases represents a significant global health challenge, necessitating early and accurate detection. This study aimed to develop and validate a unified deep learning framework capable of classifying nine distinct retinal conditions from color fundus photographs. Methods: A publicly available dataset comprising 5,318 fundus photographs across nine diagnostic categories was utilized. Images underwent standardized preprocessing, normalization, and comprehensive data augmentation to mitigate class imbalance. Three advanced deep learning architectures ResNet-151, EfficientNetV2, and a YOLOv11-based classifier were implemented, leveraging transfer learning from ImageNet. Model performance was rigorously evaluated using accuracy, precision, recall, F1-score, and AUC metrics on an independent hold-out test set. Results: The YOLOv11-based classifier achieved the highest overall accuracy (90.2%), macro-averaged recall (90.5%), precision (90.7%), and F1-score (90.4%), significantly outperforming both EfficientNetV2 and ResNet-151 models. ROC analysis confirmed superior discriminative capability (AUC=0.93) for the YOLOv11-based model, with statistically significant improvements validated through DeLong and McNemar’s tests (p<0.01). Despite robust overall performance, specific misclassifications occurred between optically similar conditions, such as optic disc edema and glaucoma, highlighting ongoing diagnostic challenges. Conclusions: This study demonstrated that a unified YOLOv11-based deep learning framework can accurately perform multi-class retinal disease classification from fundus photographs. The proposed model offers significant potential as an AI-driven screening tool, effectively addressing the limitations of traditional single-disease diagnostic systems, and providing a critical step toward improved diagnostic workflows and patient outcomes in ophthalmology.