基于CT的影像组学联合血清甲胎蛋白水平预测肝癌病理分级的价值

Value of CT-based radiomics combined with serum AFP level in predicting pathological grading of liver cancer

目的:分析基于CT的影像组学联合血清甲胎蛋白(AFP)水平预测肝癌病理分级的价值。方法:收集在医院治疗的113例经组织病理学确诊为肝细胞癌(HCC)的患者,根据患者组织病理分级将其分为高分化组(38例)和低分化组(75例),采用酶联免疫吸附法检测患者血清AFP水平。所有患者术前均行多层螺旋CT(MSCT)检查,根据最小信息准则(AIC)逐步法,分别联合CT平扫期、动脉期、门静脉期和延迟期四期相的影像组学标签及临床特征建立4个影像组学Logistic回归模型,分析CT及AFP对肝癌患者的影响因素。绘制受试者特征(ROC)曲线,分析临床特征及CT预测肝癌病理分化程度的效能。结果:高分化组患者AFP水平明显低于低分化组,其差异有统计学意义(t=4.764,P<0.01)。各期相影像组学标签对肝癌病理分化程度预测效能中门静脉期的ROC曲线下面积(AUC)为0.745,明显高于平扫期、动脉期和延迟期的0.648、0.698和0.679。高分化组四期相的影像组学评分(Rad-score)中位数明显低于同期低分化组,其差异有统计学意义(t=3.165,t=4.528,t=3.465,t=3.985;P<0.01)。AFP在鉴别肝癌病理分化程度中AUC为0.715,是独立预测因子。影像组学标签联合临床特征预测效能中,门静脉期的AUC为0.826,能较好预测肝癌病理分化程度。结论:基于CT的影像组学标签是肝癌病理分化程度的独立预测因子,其中基于门静脉期的CT图像纹理分析预测效能较好,CT影像组学联合AFP水平能够有效预测HCC的病理分级。

Objective:To analyze the value of computer tomography(CT)-based radiomics combined with serum alpha fetoprotein(AFP)level in predicting the pathological grading of liver cancer.Methods:113 patients with hepatocellular carcinoma(HCC)diagnosed by histopathology were selected and they were divided into high differentiation group(38 cases)and low differentiation group(75 cases)according to the histopathological grading.The serum AFP level was detected by enzyme-linked immunosorbent assay.All patients underwent multi-slice spiral computed tomography(MSCT)examination before surgery.According to the minimum information criterion(AIC)step-by-step method,the radiomics labels of CT plain scan phase,arterial phase,portal vein phase and delayed phase were respectively combined with clinical characteristics to establish four radiomics logistic regression models for analyzing the influence factors of CT and AFP on patients with HCC.The receiver operating characteristic(ROC)curve was drawn to analyze the clinical features and the efficiency of CT in predicting the HCC pathological differentiation.Results:The AFP level of high differentiation group was significantly lower than that of low differentiation group(t=4.764,P<0.01).The area under curve(AUC)of ROC cure of portal phase was 0.745,which was significantly higher than that of plain scan phase(0.648),arterial phase(0.698)and delayed phase(0.679).The median radiomics scores(Rad-score)of four phases of high differentiation group were significantly lower than those of low differentiation group(t=3.165,t=4.528,t=3.465,t=3.985,P<0.01),respectively.The AUC of AFP was 0.715,which was an independent predictor in identifying the degree of HCC pathological differentiation.In predicting efficiency of radiomics labels combined with clinical characteristics,the AUC of portal vein phase was 0.826,which could better predict the degree of HCC pathological differentiation.Conclusion:CT-based radiomics label is an independent predictor of the degree of HCC pathological differentiation,and the texture analysis of CT image based on portal vein phase has better prediction efficiency.CT radiomics combined with AFP level can effectively predict the HCC pathological grading.