CT影像组学预测肝硬化患者高危食管静脉曲张的价值

The Value of CT Radiomics in PredictingHigh Risk Esophageal Varices in Patientswith Liver Cirrhosis

摘要: 目的：构建肝硬化患者高危食管静脉曲张的无创预测模型。方法：回顾性分析190例确诊为肝硬化的患 者，收集肝硬化患者入院后的血清学指标及胃镜检查结果，按7:3的比例随机分为训练集(n = 133)和验 证集(n = 57)。采集研究人群的门脉期腹部强化CT图像，选脾门区作为感兴趣区，提取其影像组学特征， 应用特征相关性分析、最小绝对收缩和选择算子(LASSO)回归分析筛选特征，建立影像组学标签 (Radiomics signature, RS)并计算RS评分Rad-score；采用Logistic回归分析构建临床模型，联合 Rad-score、血清学指标及影像学指标构建结合模型。使用受试者工作特征曲线(ROC)曲线分别评估三种 模型的预测效能，曲线下面积(AUC)的比较采用Delong检验，通过决策曲线(DCA)分析评价结合模型的 临床实用性。结果：构建的影像学模型、临床模型、结合模型在训练集中的AUC分别为0.94、0.92、0.96， 在验证集中的AUC分别为0.79、0.91、0.94。在诊断效能最优的结合模型中，Rad-score及胃左静脉直径 是高危食管静脉曲张的风险因子；决策曲线显示结合模型具有良好的临床实用性。结论：本研究基于影 像组学特征、血清学指标及影像学指标构建的模型有助于预测肝硬化高危食管静脉曲张患者。

Abstract: Objective: To construct a noninvasive prediction model for high-risk esophageal varices in patients with liver cirrhosis. Methods: A retrospective analysis was performed on 190 patients diagnosed with liver cirrhosis. Serological indexes and gastroscopy results of patients admitted to hospital were collected and randomly divided into a training set (n = 133) and a validation set (n = 57) in a ratio of 7:3. The enhanced abdominal CT images of the study population at portal pulse stage were collected, and the splenic portal region was selected as the region of interest to extract its image omics features. Feature correlation analysis, minimum absolute contraction and selection operator (LASSO) regression analysis were used to screen the features. Radiomics signature (RS) was established and RS score Rad-score was calculated. Logistic regression analysis was used to construct the clinical model, combined with Rad-score, serological indicators and imaging indicators to construct the combined model. The predictive efficacy of the three models was evaluated using receiver operating characteristic curve (ROC) curves. The area under the curve (AUC) was compared using Delong test. The clinical practicability of the combined models was evaluated by decision curve (DCA) analysis. Results: The AUC of the constructed imaging model, clinical model and combination model were 0.94, 0.92 and 0.96 in the training set, and 0.79, 0.91 and 0.94 in the validation set, respectively. In the optimal combination model, Rad-score and left gastric vein diameter were risk factors for high-risk esophageal varices. The decision curve shows that the combined model has good clinical practicability. Conclusions: In this study, the model built based on the image omics characteristics, serological indicators and imaging indicators is helpful to predict the patients with esophageal varices at high risk of cirrhosis.