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

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 sethoxybenzyl 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 = 14) (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 Tl 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.