Value of MRI radiomics model in predicting hepatocellular carcinoma associated with hepatitis B virus infection

LI Siqi; YANG Cunxia; WU Chunmei; CUI Jingjing; YIN Xiaoping

doi:10.19300/j.2025.L21318

Abstract：Objective To explore the value of a multi-sequence MRI radiomics model in predicting hepatocellular carcinoma(HCC)associated with hepatitis B virus(HBV)infection.Methods A retrospective study included 147 HCC patients hospitalized for treatment and 20 HCC patients from the Cancer Imaging Archive(TCIA)public database who underwent liver resection.Patients were divided into HBV infection(HBV)and non-HBV infection(n-HBV)groups.The hospital patients were randomly assigned into a training set(102 cases)and a testing set(45 cases)in a 7∶3 ratio.The TCIA patients were used as an independent validation set.Clinical data,preoperative T2-weighted imaging(T2WI),diffusion-weighted imaging(DWI),and contrast-enhanced MRI data were collected.Two radiologists manually delineated the regions of interest(ROIs)of the lesions.Five-fold cross-validation,Pearson correlation coefficients,and the least absolute shrinkage and selection operator(LASSO)algorithm were used to extract and select radiomics features.Support vector machine(SVM)and logistic regression(LR)classifiers were employed to construct radiomics models based on T2WI,DWI,contrast-enhanced MRI,and multi-modal MRI(mpMRI:T2WI+DWI+contrast-enhanced MRI).Receiver operating characteristic(ROC)curves were used to evaluate the prediction performance of the models,and the area under the curve(AUC),sensitivity,specificity,and Brier score were calculated.The generalization ability of the best model was validated in the independent validation set.Results In both the training and testing sets,the mpMRI models based on SVM and LR classifiers performed well in predicting HCC associated with HBV infection(SVM:AUC values of 0.991 and 0.941;LR:AUC values of 0.993 and 0.936).In the testing set,the SVM model performed better than the LR model,with higher sensitivity in the SVM-based model and higher specificity in the LR-based model.The SVM-mpMRI model showed the best diagnostic performance,with an AUC of 0.792 in the independent validation set.Calibration curve analysis showed that the mpMRI model based on the LR classifier had better curve fitting to the actual curve than the model based on the SVM classifier(Brier scores of 0.006 and 0.069,respectively).Conclusions The multi-sequence MRI radiomics model can assist in non-invasively predicting whether HCC patients have HBV infection,aiding clinical decision-making and providing additional information for precise HCC diagnosis.