首页|期刊导航|现代信息科技|基于深度学习的膝关节前交叉韧带MRI图像损伤的分割模型研究

基于深度学习的膝关节前交叉韧带MRI图像损伤的分割模型研究OA

Research on a Deep Learning-Based Segmentation Model for MRI Images of Anterior Cruciate Ligament Injuries in the Knee

中文摘要英文摘要

膝关节前交叉韧带(ACL)损伤的精准影像评估对临床诊疗至关重要.针对 ACL 解剖结构细长、体积微小且背景组织复杂,导致传统深度学习模型难以兼顾全局依赖与局部细节的问题,本研究提出了一种基于 SwinUNet-FPN 的自动分割模型.该模型以纯 Transformer 架构的 Swin-UNet 为骨干,利用移动窗口自注意力机制提取全局特征,并引入特征金字塔网络(FPN)模块,通过横向连接融合深层语义与浅层细节信息.在 KneeMRI 数据集上的实验结果表明,该模型在Dice相似系数(DSC)和95%豪斯多夫距离(HD95)上分别达到0.812 7和18.234 5,显著优于U-Net、ResUNet 及原始 Swin-UNet 模型.结果证实该方法能有效克服复杂背景下的分割难题,提升边缘识别精度,具有重要的临床辅助价值.

Accurate imaging evaluation of Anterior Cruciate Ligament(ACL)injuries of the knee joint is crucial for clinical diagnosis and treatment.Due to the slender anatomical structure of the ACL,its small volume,and complex surrounding tissues,traditional Deep Learning models face difficulties in balancing global dependencies with local details.This study proposes an automatic segmentation model based on SwinUNet-FPN.The model uses Swin-UNet with a pure Transformer architecture as the backbone to extract global features through a shifted window self-attention mechanism and introduces a Feature Pyramid Network(FPN)module to fuse deep semantic information with shallow details through lateral connections.Experimental results on the KneeMRI dataset show that the model achieves Dice Similarity Coefficient(DSC)and 95%Hausdorff Distance(HD95)of 0.812 7 and 18.234 5,respectively,significantly outperforming U-Net,ResUNet,and the original Swin-UNet models.The results confirm that this method can effectively overcome segmentation challenges in complex backgrounds,improve edge recognition accuracy,and provide important clinical auxiliary value.

陆振展;任文宁;何慧敏

广西医科大学 信息与管理学院,广西 南宁 530021广西医科大学第一附属医院,广西 南宁 530021广西医科大学 信息与管理学院,广西 南宁 530021

信息技术与安全科学

前交叉韧带MRI图像分割SwinUnet特征金字塔网络(FPN)深度学习

anterior cruciate ligamentMRI image segmentationSwin-UNetFeature Pyramid Network(FPN)Deep Learning

《现代信息科技》 2026 (16)

157-162,6

广西高等教育本科教学改革工程项目(2023JGZ115广西重点研发计划项目(桂科AB25069086)

10.19850/j.cnki.2096-4706.2026.16.029

评论