Research on knee multi-class cyst detection algorithm based on cascaded Transformer and YOLOv8
ZHANG Liyuan
ZHANG Chi
JIANG Zhengang
TANG Xiongfeng
Abstract：Aiming at the high similarity and blurred boundary between the cysts and intra-articular fluid and other tissues in mag-netic resonance(MR)images of knee cysts,we proposed a knee cysts lesion detection model YOLO-Cyst.Firstly,in the backbone network,a cascaded Vision Transformer module was employed to capture long-distance contextual information,thereby enhancing cyst detection accuracy.Secondly,building upon the cross-stage partial connectivity and dual fusion module of YOLOv8,a deformable large kernel attention module was introduced to enhance the model capacity for local feature extraction.Experimental results demonstrated that compared to the YOLOv8,YOLO-Cyst improved mAP50 and mAP50-95 by 5.1%,0.8%,respectively.Furthermore,compared to the Faster R-CNN and DETR,YOLO-Cyst enhanced mAP50 by 23.9%,13.0%,and mAP50-95 by 10.7%,6.8%,respectively.This algorithm can learn rich feature representation of knee cysts and enable accurate detection of cysts of different types and morphologies.
Keywords：Knee cystsObject detectionContextual informationTransformerYOLOv8
Publication Date：2025-02-28
Online Publishing Date：2026-09-11(First online date of this platform, not the publication date of the document)
Pages：9( 58-66 )