生物骨科材料与临床研究2025，Vol.22Issue(4)：40-47,53,9.DOI:10.3969/j.issn.1672-5972.2025.04.008
基于深度学习和计算机视觉方法的矢状位腰椎核磁共振成像的自动分割与定量分析
Deep automatic segmentation and quantification of sagittal lumbar MRI based on VM-Unet and computer vision
摘要
Abstract
Objective To develop an artificial intelligence model based on VM-UNet and computer vision techniques for the automatic segmentation and quantification of key anatomical regions in sagittal T2-weighted MRI of the lumbar spine.Methods Clinical and imaging data from 1 224 patients who underwent lumbar spine MRI examinations at General Hospital,Central Theater Command of PLA were collected between June 2019 and June 2024.An automatic segmentation model for sagittal T2-weighted MRI was developed using manually segmented data from experts for training.The segmentation targets included the vertebrae,intervertebral discs,and cerebrospinal fluid.The performance of the VM-UNet,U-Net,U2-Net,and nnU-Net models was evaluated using the Dice coefficient and Hausdorff distance for segmentation performance assessment.Key anatomical region quantifications were obtained through computer vision methods,and the consistency between the model and the Image J software was examined using intraclass correlation coefficients(ICC)and Bland-Altman plots.Results In terms of automatic segmentation,the VM-UNet model exhibited excellent performance,achieving Dice coefficients and Hausdorff distances of 0.981,0.974,0.952 and 9.7,12.6,20.3,respectively,for the vertebrae,intervertebral discs,and cerebrospinal fluid,surpassing other models.For automatic quantification,the ICC between VM-UNet and manual measurements by Image J demonstrated high consistency for most quantitative measurement outcomes(ICC=0.92-0.99).Conclusion This study developed an artificial intelligence model for the automatic segmentation and quantification of sagittal T2-weighted MRI of the lumbar spine.The model is reliable,efficient,and accurate,making it an effective tool for aiding radiological diagnosis and conducting epidemiological research on lumbar spine disorders.关键词
人工智能/腰椎MRI/腰背痛/计算机视觉/模型Key words
Artificial intelligence/Lumbar spine MRI/Lumbago/Computer vision/Model分类
医药卫生引用本文复制引用
雷昌宇,朱凌,韩尧政,葛晨光,江剑峰,康辉..基于深度学习和计算机视觉方法的矢状位腰椎核磁共振成像的自动分割与定量分析[J].生物骨科材料与临床研究,2025,22(4):40-47,53,9.基金项目
湖北省医学青年拔尖人才项目[鄂卫通(2019)48号] （2019）
湖北省卫健委面上科研项目(WJ2023M091,WJ2025M133) （WJ2023M091,WJ2025M133）
