【摘要】 目的 评估基于CT影像组学结合机器学习模型术前预测食管胃交界处腺癌(AEG)人表皮生长因子受体2(HER2)状态的价值。方法 回顾性分析101例经术后病理证实的AEG患者，按7∶3比例将其分为训练集(n=70)和验证集(n=31)。基于门静脉期增强CT提取AEG影像组学特征，以最小绝对值选择与收缩算子回归模型针对训练集数据筛选最佳影像组学特征，并建立影像组学标签。采用多因素logistic回归分析筛选AEG HER2状态的独立预测因子，通过支持向量机(SVM)ML算法分别构建影像组学SVM模型和影像组学-临床联合SVM模型。应用受试者工作特征(ROC)曲线，计算相应曲线下面积(AUC),评估模型预测AEG HER2状态的效能，并比较其AUC差异。结果 101例AEG中，HER2(+)46例，HER2(-)55例。影像组学SVM模型预测训练集HER2状态的AUC为0.86,在验证集为0.78。多因素logistic回归分析显示，T分期及Rad-score为HER2状态的独立预测因子，以其建立影像组学-临床联合SVM模型，预测训练集的AUC为0.91,在验证集为0.87。影像组学SVM模型与影像组学-临床联合SVM模型预测在训练集及验证集AEG HER2状态的AUC差异均无统计学意义(Z=-2.03、-1.25,P=0.42、0.20)。结论 基于CT影像组学的SVM模型有助于术前预测AEG HER2状态，其预测效能与影像组学-临床联合SVM模型相当。更多还原

【Abstract】 Objective To explore the value of CT radiomics combined with machine learning model for preoperative predicting human epidermal growth factor receptor-2(HER2) status of adenocarcinoma at esophagogastric junction(AEG). Methods Data of 101 patients with AEG confirmed by postoperative pathology were retrospectively analyzed. The patients were divided into training set(n=70) and validation set(n=31) at the ratio of 7∶3. The radiomics features were extracted based on portal phase enhanced CT, the least absolute shrinkage and selection operator regression method was used to select the best radiomics features in training set, and then the radiomics signatures were established. Multivariate logistic regression was used to screen the independent predictors of HER2 status of AEG. Support vector machine(SVM) algorithm was used to construct the radiomics SVM model and radiomics-clinical combined SVM model, respectively. The receiver operating characteristic(ROC) curves were drawn to evaluate the efficacies of the models in predicting AEG HER2 status, and the areas under the curve(AUC) were calculated and compared. Results Among 101 AEG patients, 46 were HER2(+) and 55 were HER2(-). AUC of the radiomics SVM model for HER2 state prediction was 0.86 and 0.78 in training cohort and in validation set, respectively. Multivariate logistic regression analysis showed that T stage and Rad-score were independent predictors for HER2 status of AEG, and then were used to establish radiomics-clinical combined SVM model. AUC of radiomics-clinical combined SVM model in training set was 0.91, while in validation set was 0.87. There was no significant difference of AUC between radiomics SVM model and radiomics-clinical combined SVM model(Z=-2.03,-1.25, P=0.42, 0.20) for predicting AEG HER2 status in training set nor in validation set. Conclusion Radiomics SVM model based on CT was helpful for preoperative prediction of HER2 status of AEG, which had predictive efficacy similar to that of combining radiomics-clinical SVM model.更多还原