关键词: 肝细胞癌, 微血管侵犯, 影像组学, 计算机断层扫描

Abstract: Objective To investigate the value of single or different phases of contrast-enhanced multi-slice spiral CT（MSCT）in different volumetric regions of interest（ROI）to preoperatively predict the state of microvascular invasion in primary hepatocellular carcinoma（HCC）based on radiomics methods．Methods A total of 88 patients with HCC confirmed by surgical pathology who underwent preoperative MSCT quadruple-enhanced scan were retrospectively recruited,including 47 MVI-positive patients and 41 MVI-negative patients．The ROI was manually delineated slice-by-slice in the early arterial phase,late arterial phase,portal venous phase,and equilibrium phase of enhanced MSCT images to obtain the volume of tumor VOI（Vt）,and then Vt was expanded by 10 mm through the computer expansion algorithm automatically to obtain the volume of tumor and peritumor（Vt+Vp）．Pyradiomics software was used to extract radiomic features from Vt and Vt+Vp,followed by 150 discriminant models constructed with 150 feature selection methods and 10 classifiers,and then 10-fold cross-validation was used to evaluate the performance of these models．Using accuracy,sensitivity,specificity,area under the receiver operating characteristic curve（AUC）to assess model performance．The top three predictive models with the best performance were also compared．Results The radiomics model for predicting HCC MVI status in portal venous phase among quadruple-enhanced MSCT images outperformed other phases and different combinations of phases,achieving the highest AUC values of 0.768 and 0.782 in Vt and Vt+Vp respectively．In addition,the prediction performance of the radiomics model based on Vt+Vp was superior to models based on Vt．AUC value,accuracy,sensitivity,and specificity of the model with the best performance based on Vt+Vp were 0.782,0.728,0.745 and 0.705 respectively．Conclusions Radiomics models based on the portal venous phase of contrast-enhanced MSCT and tumor combined with the 10mm peritumoral area were more recommended to be employed to preoperative non-invasively predict the state of HCC MVI．

Key words: hepatocellular carcinoma, microvascular invasion, radiomics, computed tomography