基于增强CT影像组学联合临床特征预测肝细胞癌病理分化程度的应用研究

Application Study on Predicting the Degree of Pathological Differentiation of Hepatocellular Carcinoma Based on Enhanced CT Imaging Histology Combined with Clinical Features

目的 探讨增强CT影像组学模型联合临床特征对术前预测肝细胞癌(HCC)的病理分化程度的应用价值。方法 回顾性分析经手术切除或病理穿刺证实为HCC的196例患者的病理学及术前增强CT影像学资料。按照WHO标准将患者分为高分化组及中-低分化组,按照7:3比例将患者随机分为训练组(137例)和验证组(59例),保存增强CT动脉期(AP)、静脉期(VP)及延迟期(DP)影像学图像,在医准-达尔文科研平台中提取并筛选各期图像的影像组学特征,应用“最大值归一化法、最优特征筛选、最小绝对收缩和选择算子(LASSO)回归”进行降维、筛选,构建动脉期+静脉期、动脉期+延迟期、静脉期+延迟期及三期联合的影像组学模型。应用单因素及多因素分析方法从临床资料中筛选具有统计学差异的危险因素并建立临床模型。用逻辑回归(Logistic)方法分析影像组学模型、临床模型及影像组学联合临床模型,使用受试者工作特征曲线(ROC)评估各模型预测肝癌病理分化程度的效能,并计算曲线下面积(AUC)、准确度、特异度及敏感度等。结果 1.在影像组学模型中,三期联合模型预测肝癌病理分化程度效能最佳,训练组及验证组曲线下面积(AUC)分别为0.877及0.801。2.单因素及多因素分析结果显示AFP (P=0.010)及A LT(P=0.024)最终为预测HCC病理分化程度的危险因素。临床特征AFP及ALT构建临床模型,训练组及验证组曲线下面积(AUC)分别为0.695及0.816。3.联合模型的预测效能优于各期像影像组学模型及临床模型。训练组及验证组曲线下面积(AUC)分别为0.899及0.890。结论 基于增强CT影像组学模型联合临床模型能够准确地预测肝细胞癌(HCC)的病理分化程度。

Objective To investigate the application value of enhanced CT imaging histological model combined with clinical features for preoperative prediction of pathological differentiation of hepatocellular carcinoma(HCC).MethodsPathological and preoperative enhanced CT imaging data of 196 patients with HCC confirmed by surgical resection or pathologic puncture were retrospectively analyzed.The patients were divided into high differentiation group and middle-low differentiation group according to WHO standards,and randomly divided into training group(137 cases)and validation group(59 cases)according to the ratio of 7:3.The arterial-phase(AP),venous-phase(VP),and delayed-phase(DP)imaging images of the enhanced CT were preserved,and the imaging histological features of the images of the various phases were extracted and screened in the MediAll-In-Darwin scientific research platform,and the"maximum normalization method,optimal feature screening,minimum absolute contraction,and delayed-phase imaging"were applied."We extracted and screened the imaging features of each phase in the Medical-Quasi-Darwin Research Platform,and applied the maximum normalization method,optimal feature screening,minimum absolute shrinkage and selection operator(LASSO)regression for dimensionality reduction and screening,and constructed the imaging histological models of arterial+venous,arterial+delayed,venous+delayed,and the three-phase combination.Single-factor and multifactor analysis methods were applied to screen statistically different risk factors from clinical data and establish clinical models.Imaging histology models,clinical models and combined imaging histology clinical models were analyzed by logistic methods,and the efficacy of each model in predicting the degree of pathological differentiation of hepatocellular carcinoma was evaluated by using the subject operating characteristic curve(ROC),and the area under the curve(AUC),accuracy,specificity and sensitivity were calculated.Results1.Among the imaging models,the three-phase combined model had the best efficacy in predicting the degree of pathological differentiation of hepatocellular carcinoma(HCC),with the area under the curve(AUC)of 0.877 and 0.801 in the training and validation groups,respectively.2 The results of the unifactorial and multifactorial analyses showed that AFP(P=0.010)and ALT(P=0.024)were ultimately the risk factors for the prediction of the degree of pathologic differentiation of HCC.Clinical models were constructed with clinical features AFP and ALT,and the area under the curve(AUC)was 0.695 and 0.816 in the training and validation groups,respectively.3.The predictive efficacy of the combined model was better than that of the imaging histology model and the clinical model in all phases.The area under the curve(AUC)was 0.899 and 0.890 in the training and validation groups,respectively.Conclusion Enhanced CT image-based histologic models combined with clinical models can accurately predict the degree of pathologic differentiation in hepatocellular carcinoma(HCC).