基于增强CT影像组学列线图预测单发肝细胞癌GPC3表达状态

Value of enhanced CT radiomics nomogram in predicting GPC3 expression in single hepatocellular carcinoma

目的 探讨增强CT影像组学列线图在鉴别单发肝细胞癌(HCC)磷脂酰肌醇蛋白聚糖3(GPC3)表达中的价值。方法 回顾性收集来自2个医疗机构共152例单发HCC病人的临床及影像资料,所有病人均行上腹部增强CT扫描并记录GPC3表达水平。天津市第一中心医院的106例病人资料作为训练集(GPC3阳性83例、阴性23例),天津医科大学肿瘤医院的46例病人资料作为验证集(GPC3阳性35例、阴性11例)。对所有病人术前1个月内增强CT影像进行影像组学特征提取。在训练集中,对所有影像组学特征进行降维并得到最优子集,计算影像组学评分(Radscore);比较GPC3阳性组和阴性组间临床资料[包括血清甲胎蛋白(AFP)、糖类抗原199(CA199)等]的差异,将差异有统计学意义的指标进行二元logistic回归分析,获得GPC3阳性的独立预测因素。将获得的临床信息及Radscore分别建立临床列线图、影像组学列线图及联合列线图。采用受试者操作特征曲线下面积(AUC)分析各列线图对GPC3表达状态的预测能力,采用DeLong检验比较各列线图间的诊断效能,并用决策曲线分析评估列线图的临床价值。使用验证集数据对列线图预测效能进行验证。结果 二元logistic回归显示血清AFP、CA199、Radscore是GPC3阳性的独立危险因素[优势比(OR)分别为8.503、1.090、13 300.044,均P<0.05]。校准曲线显示联合列线图对GPC3阳性表达的预测概率与实际概率一致性良好。训练集中,联合列线图的AUC (0.918)高于影像组学列线图(0.842)和临床列线图(0.787)(均P<0.05),联合列线图的敏感度最高,而临床列线图的特异度最高;验证集中,联合列线图的AUC(0.896)高于影像组学列线图(0.726)和临床列线图(0.803)(均P<0.05),联合列线图的敏感度和特异度均最高。决策曲线分析显示当阈值概率处于16%~86%时,联合列线图的临床净获益高于临床列线图和影像组学列线图。结论 基于增强CT的影像组学列线图可以术前鉴别单发HCC GPC3阳性和阴性表达,联合列线图进一步提高了预测效能。

Objective The purpose of this study was to investigate the value of the enhanced CT radiomics nomogram model in differentiating positive and negative glypican-3(GPC3) expression in single hepatocellular carcinoma(HCC).Methods We retrospectively collected the clinical and imaging data of 152 patients with single HCC from 2 medical institutions.All patients underwent enhanced upper abdominal CT scan and recorded GPC3 expression level.Data of 106patients from Tianjin First Central Hospital were used as the train cohort(83 GPC3 positive patients and 23 GPC3 negative patients),and data of 46 patients from Tianjin Medical University Cancer Institude and Hospital were used as the validation cohort(35 GPC3 positive patients and 11 GPC3 negative patients).The contrast-enhanced CT images of all patients within 1 month before surgery were used to extract radiomics features.In the training set,we reduced radiomics features and obtain the optimal subset,and calculate Radscore for each patient.We compared the clinical information(including AFP,CA199 and so on) between the GPC3 positive and negative groups,then those parameters with statistically significant differences were fed into a binary logistic regression analysis to obtain independent predictors of positive GPC3 expression in HCC.The obtained clinical information and Radscore were used to establish clinical nomogram,radiomics nomogram,and combined nomogram,respectively,then area under the receiver operating characteristic curve(AUC) was used to analyze the prediction ability of the nomograms for hepatocellular carcinoma GPC3 expression status.DeLong test was used to compare the differences of diagnostic efficacy between the nomograms.Decision curve analysis(DCA) was used to evaluate the clinical value of the combined nomogram model.Furthermore,we conducted validation cohort for the prediction accuracy of the nomograms.Results Binary logistic regression showed serum AFP,CA199,and Radscore were independent risk factors for positive expression of GPC3(OR values were 8.503,1.090,and 13 300.044,respectively,P<0.05).The bias-corrected curve showed that the predicted probability of GPC3 positive expression was in good agreement with the actual probability.In the train cohort,the AUC of combined nomogram(0.918) was higher than that of radiomics nomogram(0.842) and clinical nomogram(0.787)(all P<0.05),and the sensitivity of combined nomogram was the highest,while the specificity of clinical nomogram was the highest.In the validation cohort,the AUC of the combined nomogram(0.896) was higher than that of the radiologic nomogram(0.726) and the clinical nomogram(0.803)(all P <0.05),and the sensitivity and specificity of the combined nomogram were the highest.The results of DCA showed that when the threshold probability was in the range of 16%-86%,the clinical net benefit value of the combined nomogram was higher than that of the clinical nomogram and the radiomics nomogram.Conclusions The enhanced CT radiomics nomogram model can distinguish the positive from negative GPC3expression of single HCC before surgery,and the combined nomogram can further improve the diagnostic efficiency.