智能提取CT图像标志点以改进凸轮型股骨髋臼撞击症的评估
本研究回顾性收集两家医院2018至2022年271例腹盆部CT,开发计算机辅助检测系统,用卷积神经网络自动选取髋部区域切片并计算α角与股骨头颈偏心距。髋部区域选取准确率达99.34%,自动与手工测量α角和股骨头颈偏心距的Pearson相关系数分别为0.964和0.856。作者认为该结果有望用于未来开发筛查CT的计算机辅助检测软件,可能辅助医生评估股骨髋臼撞击症。
要点
- 卷积神经网络自动选取髋部区域切片的准确率达99.34%。
- 自动与手工测量的α角Pearson相关系数为0.964,股骨头颈偏心距为0.856。
- Bland-Altman分析确认自动测量偏倚极小。
核心结论
- 基于卷积神经网络与图像处理算法的计算机辅助检测系统可自动选取髋部区域切片并测量α角与股骨头颈偏心距,结果有望用于未来开发筛查CT的计算机辅助检测软件,可能辅助医生评估凸轮型股骨髋臼撞击症。
研究设计与样本
- 回顾性收集两家医院2018至2022年271例常规腹盆部CT检查,每家医院各配备一台不同的CT扫描仪;设计卷积神经网络从腹盆部CT图像序列中选取髋部区域切片,该网络使用80例CT扫描进行训练,按50%、20%和30%分为训练组、验证组和测试组;随后选取通过股骨头颈复合体的最佳斜位切片,用图像处理算法计算α角与股骨头颈偏心距标志点;以每侧髋手动选取和测量的最佳斜位切片及其相关参数作为参考标准。
诊断标准与关键数值
- 卷积神经网络选取髋部区域切片的准确率为99.34%;自动与手工测量参数之间的Pearson相关系数在α角为0.964、在股骨头颈偏心距为0.856。
临床应用与阅片要点
- 作者认为该结果有望用于未来开发筛查CT的计算机辅助检测软件,可能辅助医生评估凸轮型股骨髋臼撞击症;基于人工智能的方法可在常规CT图像中更快速、更一致地评估凸轮型股骨髋臼撞击症,支持更早识别并减少临床工作流程中对操作者经验的依赖。
局限性
- 原文未报告局限性。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
英文原文(摘要)· 点击展开
OBJECTIVES: Femoroacetabular impingement (FAI) with cam-type morphology is a common hip disorder that can result in groin pain and eventually osteoarthritis. The pre-operative assessment is based on parameters obtained from x-ray or computed tomography (CT) scans, namely alpha angle (AA) and femoral head-neck offset (FHNO). The goal of our study was to develop a computer-aided detection (CAD) system to automatically select the hip region and measure diagnostic parameters from CT scans to overcome the limitations of the tedious and time-consuming process of subjectively selecting CT image slices to obtain parameters.
MATERIALS AND METHODS: 271 cases of ordinary abdominopelvic CT examination were collected retrospectively from two hospitals between 2018 and 2022, each equipped with a distinct CT scanner. First, a convolution neural network (CNN) was designed to select hip region slices among abdominopelvic CT scan image series. This CNN was trained using 80 CT scans divided into 50%, 20%, and 30% for training, validation and testing groups, respectively. Second, the most appropriate oblique slice passing through the femoral head-neck complex was selected, and AA and FHNO landmarks were calculated using image-processing algorithms. The best oblique slices were selected/measured manually for each hip as ground truth and its related parameters.
RESULTS: CT hip-region selection using CNN yielded 99.34% accuracy. Pearson correlation coefficient between manual and automatic parameters measurement were 0.964 and 0.856 for AA and FHNO, respectively.
CONCLUSION: The results of this study are promising for future development of a CAD software application for screening CT scans that may aid physicians to assess FAI.
KEY POINTS: Question Femoroacetabular impingement is a common, underdiagnosed hip disorder requiring time-consuming image-based measurements. Can AI improve the efficiency and consistency of its radiologic assessment? Findings Automated slice selection and landmark detection using a hybrid AI method improved measurement efficiency and accuracy, with minimal bias confirmed through Bland-Altman analysis. Clinical relevance An AI-based method enables faster, more consistent evaluation of cam-type femoroacetabular impingement in routine CT images, supporting earlier identification and reducing dependency on operator experience in clinical workflows.
引用
- Tayyebinezhad S, Fatehi M, Arabalibeik H, et al. Intelligent extraction of CT image landmarks for improving cam-type femoroacetabular impingement assessment. Eur Radiol. 2025. PMID: 41032172. doi:10.1007/s00330-025-11901-w
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。