Radiomics Analysis Based On MR Imaging For Prediction Of PD-L2 Expression In Hepatocellular Carcinoma

Objective:To investigate the value of multi-sequence MRI radiomics for noninvasive prediction of programmed cell death 1 Ligand 2(PD-L2)expression in patients with hepatocellular carcinoma(HCC).Materials and methods:A total of 108 patients with HCC confirmed by pathology were retrospectively analyzed.Immunohistochemical analysis was used to evaluate the expression of PD-L2.3D-Slicer software was used to manually delineate volumes of interest(VOI)and extract radiomics features(including shape features,texture features,and histogram first-order statistical features)on FS-T2WI,arterial and portal venous phase images,respectively.The least absolute shrinkage and selection operator(LASSO)was performed to select the optimal radiomics features for individual or combined analysis.Multivariable logistic regression was used to construct radiomics models,and these models were validated using 5-fold cross-validation strategy.The area under the receiver characteristic curve(AUC)was used to evaluate the predictive performance of each model.Results:Among the 108 cases of HCC,50 cases had high PD-L2 expression,and 58 cases had low PD-L2 expression.The AUC values of FS-T2WI sequence,contrast-enhanced arterial phase and portal venous phase sequence imaging model to predict the expression of PD-L2 in HCC patients were 0.852,0.814 and 0.857,respectively in the training group,and 0.789,0.727,0.770,respectively in the validation group.There was no significant difference in the predictive efficacy among the single-sequence models(P>0.05).Compared with the single-sequence model,the multi-sequence combined model showed the best performance,the AUC was 0.955(95%CI:0.921-0.989)in the training group and 0.871(95%CI:0.803-0.939)in the validation group.Conclusion:The radiomics model based on multi-sequence MRI can be used to predict the expression of PD-L2 in HCC patients.The combined model had the best performance than other single-sequence models.