FFD-YOLO11: A Lightweight and High-Precision Framework for Automated Fundus Disease Detection

Abstract

<title>Abstract</title> Automated detection of ocular lesions from fundus images is of great significance for disease screening and early diagnosis. However, existing methods are often constrained by the trade-off between model accuracy and computational efficiency, particularly under resource-limited hardware conditions, where missed detections and false positives are common.In this paper, we propose FFD-YOLO11 (Fundus Disease Detection-YOLO11), a novel and efficient detection framework based on the YOLO11 architecture. The proposed model integrates the RepViT structure and Efficient Multi-scale Attention (EMA) into the backbone to enhance the representation of pathological features. Moreover, a Large Separable Kernel Attention (LSKA) mechanism is embedded in the Spatial Pyramid Pooling Fast (SPPF) module to expand the receptive field and strengthen contextual feature modeling.Furthermore, we design a Lightweight Shared Convolutional Detection Head (LSCD) and a Feature Diffusion Pyramid Network (FDPN), which effectively fuse multi-scale features while significantly reducing the model parameters. Experimental results show that FFD-YOLO11 achieves 97.4% mAP with only 2.6M parameters, outperforming the baseline by 5.1% and achieving the best performance among comparable models. Visualization analysis further demonstrates the model’s precise focus and localization of clinically critical lesion regions.Overall, FFD-YOLO11 provides a high-accuracy, lightweight, and robust detection solution suitable for clinical environments and embedded medical imaging systems, offering a new technological approach for intelligent ophthalmic diagnosis assistance.

Affiliated Institutions

• Nanjing University of Information Science and Technology CN

Related Publications