目的 本文旨在肌肉脂肪浸润的基础上，利用分层分析的方法将肌肉内部按照不同的密度范围划分成不同的亚分区，进一步研究肌肉密度改变对原位肝移植术（OLT）后并发症（Clavien-Dindo≥Ⅲ）的影响。 方法 回顾性分析2013年5月—2020年9月于吉林大学第一医院行OLT的145例患者，以患者腰3椎体水平最大层面的CT平扫图像作为原始数据，利用Neusoft Fatanalysis软件对图像进行相关肌肉参数的测量。符合正态分布的计量资料组间比较采用成组t检验；不符合正态分布的组间比较采用Mann-Whitney U秩和检验。计数资料组间比较采用χ²或Fisher检验。利用RIAS软件进行临床特征提取及分析建模，分别建立逻辑回归（LR）、支持向量机（SVM）、随机森林（RFC）3种机器学习模型，并绘制不同模型的受试者操作特征曲线（ROC曲线）、校正曲线、决策分析曲线，计算ROC曲线下面积（AUC）、灵敏度、特异度、精确率、F1分数、准确率。 结果 采用肌肉分层分析前的7种临床特征建立LR-C、SVM-C、RFC-C 3种机器学习模型，其中RFC-C模型测试集的AUC值为0.803、灵敏度0.588，特异度0.778。采用肌肉分层分析后的16种临床特征建立的LR-CS、SVM-CS、RFC-CS模型中，LR-CS及SVM-CS模型测试集的AUC值较高，均为0.852，灵敏度分别为0.765、0.706，特异度分别为0.889、0.926，通过对比肌肉分层分析前后各模型测试集的AUC、灵敏度、特异度、精确率、F1分数、准确率后发现，肌肉分层分析后预测模型的参数均有所提升。通过对比各预测模型的决策分析曲线和校正曲线，发现LR-CS及SVM-CS模型对于预测OLT患者术后并发症（Clavien-Dindo≥Ⅲ）具有良好效能。 结论 在肌肉脂肪浸润的基础上，利用分层分析的方法将肌肉内部按照不同的密度划分成不同子区，对于OLT患者术后并发症有一定预测价值。

Objective To divide the muscle into different subzones according to different density ranges using the stratified analysis on the basis of myosteatosis， and to investigate the effect of muscle density changes on complications （Clavien-Dindo grade ≥Ⅲ） after orthotopic liver transplantation （OLT）. Methods A retrospective analysis was performed for the medical records of 145 patients who underwent OLT in The First Hospital of Jilin University from May 2013 to September 2020， and with the plain CT scan images of the largest level of lumbar 3 vertebrae of each patient as the original data， Neusoft Fatanalysis software was used to measure related muscle parameters. The independent-samples t test was used for comparison of normally distributed continuous data between two groups， and the Mann-Whitney U test was used for comparison of non-normally distributed continuous data between two groups. The chi-square test or Fisher test was for comparison of categorical data between two groups. RIAS software was used to extract clinical features and perform analysis and modeling， and three machine learning models of logistic regression （LR）， support vector machine （SVM）， and random forest （RFC） were constructed. The receiver operating characteristic （ROC） curve， the calibration curve， and the decision curve were plotted for each model to calculate the area under the ROC curve （AUC）， sensitivity， specificity， precision， F1 score， and accuracy. Results The three machine learning models of LR-C， SVM-C， and RFC-C were established based on the 7 clinical features before muscle stratification analysis， among which the RFC-C model had an AUC of 0.803， a sensitivity of 0.588， and a specificity of 0.778 in the test set. Among the models of LR-CS， SVM-CS， and RFC-CS established based on the 16 clinical features after muscle stratification analysis， the LR-CS and SVM-CS models had an AUC of 0.852 in the test set， with a sensitivity of 0.765 and 0.706， respectively， and a specificity of 0.889 and 0.926， respectively. Comparison of the AUC， sensitivity， specificity， precision， F1 score， and accuracy of each model in the test set before and after muscle stratification analysis showed that there were improvements in the parameters of the predictive model after muscle stratification analysis. Comparison of the decision curves and calibration curves of each predictive model showed that the LR-CS and SVM-CS models had good efficacy in predicting postoperative complications （Clavien-Dindo grade≥Ⅲ） in OLT patients. Conclusion On the basis of myosteatosis， the division of the muscle into different subzones according to different densities using the stratified analysis has a certain value in predicting postoperative complications in patients with OLT.