| Objective:To investigate the value of a radiomic model based on Gd-EOB-DTPA enhanced MRI images in preoperative prediction of microvascular invasion（MVI）staging in hepatocellular carcinoma（HCC）.Methods:A total of 165 patients with pathologically confirmed HCC（M0:M1:M2=89:53:23）at the First Hospital of Jilin University from December 2017 to June 2020 were retrospectively enrolled（training cohort,n=115;testing cohort,n=50）.The enhanced MR examination images of the patients within two weeks before curative hepatectomy were collected.The regions of interest（ROIs）of the hepatic lesions were manually delineated in the T2-FS,L-AP,PVP,HBP,and DWI（b=800 s/mm~2）using 3D-Slicer software.Rise software was applied for extracting the radiomic features,and ANOVA analysis and LASSO regression were used for dimensionality reduction and selection of the extracted features.For constructing the MVI classification model,this study adopted the modelling idea of a one-vs-one parallel three classification model,a multi-classification model by using the logistic regression with reduced dimensional radiomics features to predict the grade of MVI（M0,M1,M2）.The clinical variables related to MVI grading,radiomics features and radiomics lasso score were fused to form a clinicoradiological fusion model.Finally,the radiomics model and clinicoradiological fusion model were evaluated and compared.Results:The accuracy of the radiomics model in the test set was 0.74,the precision was 0.71,the Hemming distance was 0.26,the Jaccard similarity coefficient was 0.74,the F1 score was 0.72,and the recall was0.74;the AUC of the radiomics model was 0.83.The accuracy of the clinicoradiological fusion model in the test set was 0.72,the precision was 0.71,the Hemming distance was 0.28,Jaccard similarity coefficient of 0.72,F1 score of 0.71,and recall of 0.74;the AUC of the fusion model in the test set was 0.82.Conclusion:The three categories radiomics model of MVI was superior to the clinicoradiological fusion model in this study.It has a certain value in preoperative prediction of the occurrence and grading of microvascular invasion in HCC patients. |
