Prediction Of The Refractory To Initial TACE In Hepatocellular Carcinoma Based On Radiomic Analysis Of Contrast-enhanced CT:A Multiinstitutional Study

Background and Objective: We aim to develop a nomogram model integrating clinical and radiomic features for predicting patients who are refractory to initial TACE(RIT),the progression of vascular invasion or extrahepatic spread.Methods: In this retrospective study,256 patients with unresectable HCC who underwent TACE between April 2007 and June 2018 at three institutions were included.1218 radiomic features were exacted from portal venous phase CT scans with an opensource Pyradiomics package.In the training cohort(institution Ⅰ and Ⅱ,n=136),a machine learning algorithm linear discriminant analysis and analysis of variance was used for building a radiomics score.A clinical model for predicting RIT was developed with clinical factors selected by multivariate logistic regression analysis.Clinical model and radiomics-clinical combined model were evaluated and validated in the independent external testing cohort(institution Ⅲ,n=120).Results: In the training cohort,25% of the patients(n=34)met the RIT criteria,whereas22% of the patients(n=26)confirmed in the testing cohort.The median overall survival(OS)of the RIT group was shorter than the non-RIT group(13.5 vs.26.9 months,P<0.01;10.8 vs.32.8 months,P<0.01).Tumor number,bilirubin,tumor capsule,and radiomics score were independent risk factors of RIT.The clinical-radiomics combined model showed superior predictive performance compared to the clinical model in the training cohort(Area under the curve(AUC)0.911[95%CI: 0.862-0.961] vs.0.772[95%CI: 0.676-0.868];P<0.01),and better performance with no statistical significance in the testing cohort(AUC 0.847[95%CI: 0.775-0.919] vs.0.746[0.641-0.852];P=0.056).The combined model shows good agreement between prediction and observation in calibration curves(P>0.05).Decision curve analysis shows larger net benefit of the combined model that further confirmed its clinical usefulness.Conclusions: The model incorporating both clinic-radiological risk factors and radiomic features based on portal phase CT could preoperatively predicting patients with refractory to initial TACE in HCC and may help to improve clinical decision making.