基于临床及CT影像组学模型早期预测结直肠癌肝转移的化疗反应

Early prediction of chemotherapy response in colorectal cancer liver metastasis based on clinical and CT radiomics models

目的探讨临床、病理联合CT影像组学特征构建的联合模型早期预测结直肠癌肝转移(CRLM)化疗反应的价值。方法回顾性收集169例CRLM病人的临床、病理、增强CT影像资料。每例病人随机抽取1个肝转移灶,共169个。根据实体肿瘤疗效评价标准(RECIST)将病灶分为2组,即化疗有反应组(75个)和无反应组(94个)。按7∶3比例将病灶随机分为训练集(118个)和验证集(51个)。提取基线门静脉期CT影像中病灶的影像组学特征,采用Pearson相关系数、Select percentile单因素分析和最小绝对收缩和选择算子(LASSO)筛选最优影像组学特征。使用Logistic回归分类器构建影像组学模型并计算影像组学评分(Radscore)。采用t检验、Mann-Whitney U检验及卡方检验筛选2组间差异有统计学意义的临床和病理特征,并与Radscore结合分别构建临床-病理模型和联合模型。采用受试者操作特征(ROC)曲线、校准曲线、Hosmer-Lemeshow检验评价模型的预测效能、校准度及拟合度,并以联合模型预测指标构建列线图。结果共筛选出9个最优影像组学特征,3个临床和病理特征(肝转移类型、癌胚抗原及RAS基因)。在训练集和验证集中,联合模型预测肝转移灶化疗反应的效能(AUC=0.896,0.798)与影像组学模型(AUC=0.895、0.786)近似,但差异无统计学意义(均P>0.05)。在训练集中,联合模型预测肝转移灶化疗反应的效能高于临床-病理模型(P<0.05),而验证集中则与临床-病理模型差异无统计学意义(P>0.05)。校准曲线、Hosmer-Lemeshow检验显示联合模型列线图的校准度和拟合度均良好。结论基于化疗前CT影像组学特征构建的影像组学模型可以早期预测CRLM的化疗反应,联合临床、病理特征可略微提高模型的预测效能。

Objective To investigate the value of a combined model based on clinical,pathological,and CT radiomics features in early predicting chemotherapy response in colorectal cancer liver metastasis(CRLM).Methods Clinical,pathological,and contrast-enhanced CT imaging data of 169 patients with CRLM were retrospectively collected.One liver metastatic lesion was randomly selected from each patient,totaling 169 lesions.Lesions were divided into two groups according to Response Evaluation Criteria in Solid Tumors(RECIST):chemotherapy responsive group(75 lesions)and chemotherapy non-responsive group(94 lesions).Lesions were randomly divided into training set(118 lesions)and validation set(51 lesions)in a 7∶3 ratio.Radiomics features of lesions in baseline portal venous phase CT images were extracted.Optimal radiomics features were selected using Pearson correlation coefficient,Select Percentile univariate analysis,and Least Absolute Shrinkage and Selection Operator(LASSO).A radiomics model was constructed using logistic regression classifier,and radiomics score(Radscore)was calculated.Clinical and pathological features with statistically significant differences between the two groups were selected using t-tests,Mann-Whitney U tests,and Chi-square tests.Clinical-pathological model and combined model were constructed by integrating selected features with Radscore.The predictive performance,calibration,and goodness-of-fit of the models were evaluated using receiver operating characteristic(ROC)curve,calibration curve,and Hosmer-Lemeshow test.Nomogram were constructed based on the predictive indicators of the combined model.Results Nine optimal radiomics features,three clinical and pathological features(type of liver metastasis,carcinoembryonic antigen,and RAS gene)were selected.In the training and validation sets,the predictive efficacy of the combined model for chemotherapy response in liver metastatic lesions(AUC=0.896,0.798)was similar to the radiomics model(AUC=0.895,0.786),with no statistically significant difference(all P>0.05).In the training set,the predictive efficacy of the combined model for chemotherapy response in liver metastatic lesions was higher than the clinical-pathological model(P<0.05),while in the validation set,there was no statistically significant difference compared to the clinical-pathological model(P>0.05).Calibration curve and Hosmer-Lemeshow test showed good calibration and fit of the Nomogram of the combined model.Conclusion The radiomics model based on CT imaging features before chemotherapy can early predict chemotherapy response in CRLM.Integration of clinical and pathological features can slightly improve the predictive performance of the model.