基于多模态影像学手段构建肝癌微血管侵犯的预测模型

The predictive model of microvascular invasion in hepatocellular carcinoma established based on multimodality imaging

目的探索肝细胞癌(HCC)的微血管侵犯(MVI)发生的风险因素,实现术前、无创、准确预测MVI。方法回顾性收集2016年1月至2022年6月在西安交通大学第一附属医院确诊的150例HCC患者(183个HCC病灶)。记录患者的临床和血液学数据、病灶的灰阶超声(US)、超声造影(CEUS)、钆塞酸二钠增强磁共振(EOB-MRI)和病理学数据。根据病理学诊断分为MVI阳性组(109个)和MVI阴性组(74个)。比较两组间以上参数的差异。全部样本纳入训练集,根据多因素逻辑回归筛选出的MVI风险因素构建列线图预测模型。采用十重交叉验证法进行内部验证。结果MVI阳性组与MVI阴性组间以下指标差异有统计学意义(均P<0.05),包括有无肝硬化、血清学指标(甲胎蛋白、白蛋白、总胆红素)、US定性指标(尺寸、边界、内部回声)、CEUS定性指标[动脉相(AP)、静脉相(PP)、血管后相(PVP)病灶较周边组织的高/等/低增强]和EOB-MRI的定量指标[主要根据肝胆相、平扫的T1图像上病变和周边肝实质计算的增强后率(post率)及钆塞酸二钠率(EOB率)]。肝硬化,US的病灶尺寸、边界、内部回声,CEUS的AP、PP、PVP特征,以及EOB-MRI的EOB率、post率最终进入MVI的预测模型。训练集可获得较好的校准度和净获益率。训练集和验证集的ROC曲线下面积分别为0.981和0.961,诊断准确率分别为92.9%和85.8%。结论以多模态影像学为主要手段构建的模型显示出对MVI良好的预测性能,为无创预测MVI发生率及优化HCC患者的MVI相关治疗提供了有价值的信息。

Objective To explore the risk factors of microvascular invasion(MVI)in hepatocellular carcinoma(HCC),and to predict MVI preoperatively,non-invasively and accurately.Methods A total of 150 HCC patients(183 HCC lesions)were retrospectively collected in the First Affiliated Hospital of Xi′an Jiaotong University from January 2016 to June 2022.The clinical data and hematological data,gray-scale ultrasonography(US),contrast-enhanced ultrasonography(CEUS),enhanced magnetic resonance imaging with gadolinium ethoxybenzyl diethylenetriamine pentaacetic acid(EOB-MRI)and pathological data of these patients were recorded.According to the pathological diagnosis of MVI,the lesions were divided into MVI(+)group and MVI(-)group.The indicators between the two groups were compared.All 183 lesions were put into the training set,and the prediction model with nomogram was constructed according to the risk factors of MVI selected by multivariate Logistic regression.The internal verification was carried out by ten-fold cross-validation method.Results There were significant statistical differences in the following parameters between MVI(+)group(n=109)and MVI(-)group(n=74)(all P<0.05).These were cirrhosis,serological parameters(alpha-fetoprotein,albumin,total bilirubin),qualitative indexes of US(size,boundary,internal echo),qualitative indexes of CEUS(hyper/iso/hypovascularity of lesions in arterial phase,portal phase,and delayed phase compared with hepatic parenchyma),and quantitative indexes of EOB-MRI[post enhancement rate(post ratio)and gadolinium disodium rate(EOB ratio)]calculated mainly in terms of lesions and surrounding liver parenchyma in hepatobiliary phase and unenhanced T1 images.Finally,cirrhosis of patients,the size,boundary,internal echo of lesions in US;arterial phase(AP),portal phase(PP),post-vascular phase(PVP)features in CEUS;the EOB rate and post rate of EOB-MRI entered the prediction model of MVI.The training set exhibited good calibration and net gain rate.The areas under the ROC curve for the training set and the validation set were 0.981 and 0.961,respectively,while the diagnostic accuracy were 92.9%and 85.8%,respectively.Conclusions The model constructed mainly by multimodality imaging methods can achieve favorable predictive performance for MVI,which provides valuable ideas for noninvasively predicting the incidence of MVI and optimizing the MVI-related treatment of MVI in HCC patients.