基于改进YOLOv8的眼底微动脉瘤检测算法OA

The detection of fundus microaneurysm holds important clinical significance for screening dia-betic retinopathy in early stage.Given the relatively small target area of microaneurysm lesions and the in-terference of other fundus structures during detection,a YOLO-FDS network model based on improved YOLOv8 is proposed.First,a lesion-centered region segmentation method is applied to the dataset for lo-cal enhancement.The C2f_Star module is introduced into the backbone network,which utilizes the star operation to be more lightweight while improving the feature fusion efficiency and performance.Second,the Neck part is improved into the FDPN-DASI network,enabling features at each scale to retain detailed contextual information.Then,the network can adaptively select the fused features to improve the saliency of the target.Finally,Wise-IoU is used as the loss function to effectively improve the overall performance of the detector.Experimental results demonstrate that the proposed algorithm achieves superior detection performance on the dataset,with an mAP of 84.6%,representing a 7.1%improvement over the original model.Additionally,the precision is increased by 6.06%and the recall rate by 6.95%,while the detec-tion speed is notably faster.