The Value Of Radiogenomics In Prediction Of Prognosis Of Hepatocellular Carcinoma After Radical Hepatectomy

BackgroundPurposeFirst,to develop and validate a radiomics signature for preoperative prediction of early recurrence(? 2 years)in patients with hepatocellular carcinoma(HCC)after radical hepatectomy and apply biological function represented by genes to improve the interpretation of the predictive model of radiomics.Second,to construct radiomics signature and genomics signature for predicting recurrence-free survival(RFS),and explore whether a combined multi-omics model could improve prediction performance.MethodsPart I:the predictive model was constructed for predicting early recurrence.A total of 262 HCC patients undergoing radical hepatectomy were enrolled(training cohort=214 and validation cohort=48).Pyradiomics software extracted the radiomics features from the 3D tumour target volume of the preoperative computed tomography(CT)image.Algorithms such as propensity score matching(PSM)and least absolute shrinkage and selection operator-Logistic regression(LASSO-Logistic)were used for evaluating reproducibility,removing redundancy,reducing data dimension,and construction of a radiomics signature.Multivariate logistic regression identified independent risk factors that were used to develop a nomogram model.The predictive models were analyzed in terms of its discriminative ability,calibration and clinical usefulness both in training cohort and validation cohort.Part II:the predictive model was constructed for predicting RFS.A total of 33 eligible HCC patients from The Cancer Genome Atlas(TCGA)&The Cancer Imaging Archive(TCIA)database were enrolled.Algorithms such as LASSO-Cox regression were used to develop radiomics-based and genomics-based prognosis model respectively.Multivariate Cox regression analysis identified independent prognostic factors that were used to construct a nomogram model.The predictive performance of these three models was compared and complementary predictive performance between radiomics signature and genomics signature was also explored.Finally,a correlation between the features of radiomics signature contained in both part and the gene models clustered by Weighted gene co-expression network analysis(WGCNA)was constructed.The biological function enriched by gene enrichment analysis was used to intuitively explain the predictive performance of radiomics features.ResultsPart ?:30 features screened from 1665 radiomics features were used to develop radiomics signature.The radiomics signature was significantly related to RFS and overall survival(OS)(P<0.001).Multivariate logistic regression abalysis identified serum alpha-fetoprotein levels(AFP),tumour number,tumour size and radiomics signature as the key parameters related to early recurrence,and these factors were incorporated into the radiomics nomogram.The radiomics nomogram showed the highest area under the receiver operating characteristic curve(AUC)and a significant improvement in predictive performance as compared with the clinical nomogram in both cohort(the AUC in the training cohort:0.800 vs 0.716,P=0.001;The AUC in the validation cohort:0.785 vs 0.654,P=0.039).Its clinical usefulness was confirmed by the decision curve analysis.Part ?:The radiomics signature included 9 features screened from 1665 radiomics features and the genomics signature included 6 genes screened from 20531 genes.Univariate Cox regression analysis indicated that gender,alcohol hepatitis,AFP,LRP1B mutations,genomics signature and radiomics signature were significantly associated with RFS and OS(P<0.05).Multivariate Cox regression analysis identified genomics signature and radiomics signature as independent risk factors,which were used to establish a nomogram.The nomogram,genomics signature and radiomics signature all have good predictive performance(C-index is 0.921,0.896 and 0.866 respectively),and the nomgram is significantly better than the others(P<0.05).The analysis of GeneOntology(GO)and Kyoto Encyclopedia of Genes and Genomes(KEGG)revealed that genes in the WGCNA modules that were significantly related to radiomics features were mainly enriched in biological functional pathways such as immune response,angiogenesis,cell adhesion,and cell proliferation,which were all identified to be closely related to tumour recurrence and metastasis.ConclusionsThe radiomics signature and genomics signature may function as a prognostic biomarker for patients with HCC after radical hepatectomy.The predictive performance of clinical characteristics,radiomics features and gene data are complementary.The combined multi-omics model is better than the single-omics model in predictive performance.The radiomics features,which can act as a prognostic biomarker for HCC,may be associated with the biological function representing tumour aggressive behavior.