自动3D胸部CT肌肉分割模型鉴别正常、PRISm与COPD的多中心研究
该多中心回顾性研究纳入1660例受试者(正常616例、PRISm 646例、COPD 398例),利用TotalSegmentator自动分割胸部CT的胸肌与竖脊肌并构建支持向量机模型。结果显示3D肌肉参数模型鉴别三种肺功能状态的宏平均AUC为0.735,优于2D模型的0.646。竖脊肌密度是最具鉴别价值的肌肉参数,且从正常到PRISm再到COPD呈递减趋势。
要点
- 3D肌肉参数模型鉴别正常、PRISm与COPD的宏平均AUC达0.735,优于2D模型的0.646。
- 竖脊肌密度是区分三种肺功能状态最相关的肌肉参数。
- 3D肌肉密度从正常到PRISm到COPD呈递减趋势,提示肌少症与肺功能减退相关。
核心结论
- 基于3D肌肉参数的模型对正常、PRISm与COPD三分类的鉴别性能优于2D模型,测试集宏平均AUC为0.735。
- 3D竖脊肌密度是最具鉴别价值的肌肉参数,从正常到PRISm到COPD呈递减趋势。
研究设计与样本
- 多中心回顾性研究,共纳入1660例受试者(正常616例、PRISm 646例、COPD 398例)。
- 中位年龄66岁(IQR 60-72岁),男性1014例,按训练集1162例与测试集498例划分。
- 采用TotalSegmentator自动分割胸部CT获取3D胸肌与竖脊肌体积,并对3D与2D定量参数进行归一化。
- 分别基于归一化3D与2D肌肉参数建立支持向量机模型,采用十折交叉验证评估。
- 使用ROC曲线分析模型鉴别性能,并采用SHapley Additive exPlanations算法量化各参数重要性。
关键影像表现/征象
- 3D肌肉密度从正常到PRISm再到COPD呈递减趋势。
- 3D竖脊肌密度是三分类鉴别中最重要的肌肉参数。
- COPD女性主要表现为肌肉密度降低,男性则以肌肉体积降低更显著,尤其胸大肌体积指数。
诊断标准与关键数值
- 测试集中3D模型宏平均AUC为0.735,微平均AUC为0.733。
- 测试集中2D模型宏平均AUC为0.646,微平均AUC为0.665。
- COPD女性肌肉密度差值为-14.48至-7.79(p < 0.001)。
- COPD男性胸大肌体积指数差值为-11.18(95%CI -15.57至-5.08;p < 0.001)。
临床应用与阅片要点
- 胸部CT报告中可关注竖脊肌密度变化,密度减低可能提示肺功能受损甚至COPD。
- 3D肌肉定量分析较传统2D评估能更好区分正常、PRISm与COPD,有助于个体化综合评估。
- 不同性别COPD患者的肌肉改变模式不同,女性以密度下降为主,男性以体积减少更突出。
局限性
- 本研究为回顾性设计,存在选择偏倚可能。
- 研究未报告外部独立验证集验证模型的泛化能力。
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: To develop and validate an automated 3D chest CT muscle segmentation model using an exploratory normalization method to discriminate normal, preserved ratio impaired spirometry (PRISm), and chronic obstructive pulmonary disease (COPD) subjects.
MATERIALS AND METHODS: In this retrospective study, we used TotalSegmentator to acquire 3D pectoral muscle (PM) and erector spinae muscle (ESM) volumes from chest CT, normalizing them to height and vertebral bone volume, respectively. 2D quantitative parameters were acquired and normalized. The support vector machine models were developed based on normalized 3D and 2D muscle parameters, respectively, with ten-fold cross-validation. Receiver operating characteristic curves were used to analyze the discriminative performance of models, and the SHapley Additive exPlanations algorithm was used to quantify the importance of each parameter.
RESULTS: 1660 subjects (616 normal, 646 PRISm, 398 COPD) (median age, 66 years [IQR: 60-72 years]; 1014 males) were included and divided into training set (n = 1162), test set (n = 498). In the test set, the 3D model (macro-average AUC, 0.735; micro-average AUC, 0.733) outperformed the 2D model (macro-average AUC, 0.646; micro-average AUC, 0.665), and 3D ESM density was the most relevant muscle parameter. 3D muscle density showed a decreasing trend from normal to PRISm to COPD. In COPD, females exhibited decreased density (difference range: -14.48 to -7.79; p < 0.001), while males had a more significant decrease in volume, especially pectoralis major volume index (difference: -11.18; 95% CI -15.57, -5.08; p < 0.001).
CONCLUSIONS: The 3D muscle parameter model demonstrated higher discriminative performance than the 2D model for discriminating normal, PRISm, and COPD subjects.
KEY POINTS: Question The discriminative performance of 3D CT-based muscle quantification in subjects with impaired pulmonary function remains unclear. Findings 3D muscle analysis achieved superior discrimination of normal, PRISm, and COPD subjects, with density-related parameters showing consistent group differences, particularly ESM density. Critical relevance By comparison with conventional 2D assessment, 3D CT-based muscle quantification showed better discrimination of normal, PRISm, and COPD, which might advance the personalized precision and comprehensive assessment, including muscle alteration for subjects with impaired pulmonary function.
引用
- Wang, Zhai, Zhou, et al. Automated 3D chest CT muscle segmentation-derived model development and explanation in discrimination of normal, preserved ratio impaired spirometry, and COPD: a multicenter study. Insights Imaging. 2026. PMID: 42658393. doi:10.1186/s13244-026-02384-4
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。