CT纹理分析对于术前预测直肠癌淋巴结转移的初步研究

Texture Analysis of Preoperative CT Images for Prediction of Lymph Node Metastasis: A Preliminary Study in Patients with Rectal Cancer

目的探讨基于增强CT的纹理分析对于术前预测直肠癌淋巴结转移的应用价值。方法回顾性分析93例经术后病理证实的直肠癌患者静脉期CT图像,进行肿块全层感兴趣区(ROI)的勾画,并提取纹理特征。通过LASSO回归进行特征降维,构建影像组学评分公式。采用独立样本t检验或卡方检验对淋巴结转移组及非转移组的一般临床特征、常规CT评估及影像组学评分进行比较。采用多因素Logistic回归筛选独立预测因子,绘制受试者工作特征曲线(ROC)评估模型的预测效能,运用校准曲线评估所构建模型的预测概率与实际概率的一致性。结果单因素分析显示癌胚抗原(CEA)、常规CT评估淋巴结状态、影像组学评分对于预测淋巴结转移有统计学意义(P<0.05)。Logstic回归分析显示,常规CT评估淋巴结状态、影像组学评分是判断淋巴结转移的独立预测因子(OR值分别为3.40、2.41,P<0.05)。影像组学评分曲线下面积(AUC)、敏感度及特异度分别为0.894、0.894和0.826,常规CT评估AUC、敏感度及特异度分别为0.689、0.660和0.717,联合应用后AUC、敏感度及特异度分别为0.906、0.830和0.870。基于影像组学评分及常规CT评估淋巴结状态两个独立预测指标构建nomogram预测模型,一致性指数(C-index)为0.908。标准曲线显示nomogram预测效果与淋巴结转移的实际情况一致性较好。结论影像组学模型能够实现术前预测淋巴结转移,可为临床医师治疗方案的选择提供影像学证据。

Objective To explore the value of texture analysis based on enhanced-CT for preoperative prediction of lymph node metastasis in patients with rectal cancer. Methods 93 patients with rectal cancer confirmed by postoperative pathology were included retrospectively.All the patients underwent abdominal enhanced CT scan.Whole-tumor regions of interest(ROIs) were delineated and the texture features were extracted.LASSO regression model was used for data dimension reduction and radiomics scoring formula building.The general clinical features, CT-reported lymph node status and radiomics scores between metastatic and nonmetastatic lymph node group were compared by independent sample t-test or chi-squaretest.Multivariable Logistic regression analysis was used to choose the independent predictors.The receiver operating characteristic(ROC) curves were depicted to measure the predictive performance.The calibration curve was used to evaluate the consistency between the predicted probability and the actual probability of the model. Results Univariate analysis showed that CEA,CT-reported lymph node status and radiomics scores were statistically significant in predicting lymph node metastasis(P< 0.05).Further Logistic regression analysis indicated that CT-reported lymph node status and radiomics scores were independent predictors of lymph node metastasis(OR = 3.40,2.41,P<0.05,respectively).Based on the radimoicsmodel, the AUC was 0.894,the sensitivity was 0.894 and the specificity was 0.826.Based on the conventional CT,the AUC was 0.689,the sensitivity was 0.660 and specificity was 0.717.The combination model had an AUC of 0.906,a sensitivity of 0.830 and a specificity of 0.870.A nomogram prediction model was constructed based on radimoics scores and CT-reported lymph node status, with a C-index of 0.908.The standard curve showed that the prediction effect of nomogram was consistent with the actual situation of lymph node metastasis. Conclusion The radiomics model can predict lymph node metastasis preoperatively and provide imaging evidence for clinicians to select treatment plans.