联合CT影像组学与深度学习特征建立列线图预测食管鳞癌放疗近期疗效

To predict the short-term efficacy of radiotherapy for esophageal squamous cell cancer by nomogram based on CT radiomics and deep learning features

目的:探讨基于CT影像组学特征与深度学习特征建立列线图对食管癌放疗近期疗效的预测价值。方法:回顾性分析137例食管鳞癌患者的临床及影像资料。从CT图像中提取影像组学特征和深度学习特征。通过最小绝对收缩和选择算子方法分别对影像组学特征和深度学习特征进行降维并计算得到影像组学得分(Radscore)和深度学习得分(Deepscore)。采用多因素logistic回归分析建立预测模型,并绘制列线图。对列线图的校准度、诊断效能和临床价值进行评价。结果:筛选得到6个影像组学特征参与计算Radscore,6个深度学习特征参与计算Deepscore。多因素logistic回归结果显示Radscore、Deepscore、TNM分期为联合模型的独立预测因子。联合预测模型在训练集中预测食管鳞癌患者放疗近期疗效的曲线下面积(AUC)为0.904,高于临床模型(AUC=0.662)和影像组学模型(AUC=0.814),且AUC差异均有统计学意义(P<0.001、P=0.004)。验证集中联合模型的AUC为0.938,高于临床模型(AUC=0.644)和影像组学模型(AUC=0.852),联合模型与临床模型间AUC差异有统计学意义(P<0.001),与影像组学模型间AUC差异无统计学意义(P=0.091)。决策曲线分析发现联合预测列线图在0.1~0.9和0.97~0.99的阈值范围内表现出较好的临床实用性。结论:CT影像组学特征联合深度学习特征能较好地预测食管癌放疗近期疗效。

Objective:To explore the predictive value of nomogram based on CT radiomics and deep learning features for the short-term efficacy of radiotherapy in oesophageal cancer.Methods:The clinical and imaging data of 137 patients with esophageal squamous carcinoma were retrospectively analyzed.Radiomics and deep learning features were extracted from CT images.The least absolute shrinkage and selection operator methods were used to reduce the dimension of radiomics features and deep learning features,respectively,and the radiomics score(Radscore)and deep learning score(Deepscore)were calculated.Multivariate logistic regression analysis was used to establish a prediction model and draw a nomogram.The calibration,diagnostic efficiency and clinical value of the nomogram were evaluated.Results:Six radiomics features were selected to calculate the Radscore,and six deep learning features were selected to calculate the Deepscore.The results of multivariate logistic regression showed that Radscore,Deepscore and TNM staging were independent predictors of the combined model.The area under the curve(AUC)of the combined prediction model in the training set was 0.904,which was higher than that of the clinical model(AUC=0.662)and the radiomics model(AUC=0.814),and the differences in the AUCs were statistically significant(P<0.001 and P=0.004).In the validation set,the AUC of the combined model was 0.938,which was higher than that of the clinical model(AUC=0.644)and radiomics model(AUC=0.852).The difference in AUC between the combined model and the clinical model was statistically significant(P<0.001),while the difference in AUC between the combined model and the radiomics model was not statistically significant(P=0.091).Decision curve analysis showed that the combined prediction nomogram had good clinical practicability within the threshold ranges of 0.1~0.9 and 0.97~0.99.Conclusion:CT radiomics combined with deep learning features can better predict the short-term efficacy of radiotherapy for esophageal cancer.