The Value Of MRI In Predicting Recurrence After Radiofrequency Ablation Of Hepatocellular Carcinoma

Purpose To assess the optimal cut-off value between early recurrence and late recurrence after radiofrequency ablation(RFA)in patients with hepatocellular carcinoma(HCC),and to analysis the clinical and magnetic resonance imaging(MRI)risk factors for early recurrence and to develop a predictive nomogram.Materials and methods 119 patients with HCC who recurred after RFA in Cancer Hospital,Chinese Academy of Medical Sciences from January 2012 to December 2017 were identified.The optimal cut-off value to distinguish early and late recurrence was determined based on differences in post recurrence survival(PRS)by minimum P-value approach.The consistency of MRI feature of HCC was test by Kappa value.Clinical and radiographic risk factors for early recurrence were identified by univariate and multivariate logistic regression analysis and collinearity analysis.Predictive nomogram was developed by these risk factors and internally validated.Results The optimal cut-off value to distinguish early recurrence and late recurrence was 12 months after RFA(P=0.005).The patients were divided into early recurrence group(47 cases)and late recurrence group(72 cases).The lower quartile PRS(Q1-PRS)and lower quartile overall survival(Q1-OS)were 11.1 and 19.1 months in the early recurrence group,which were shorter than those in the late recurrence group(31.6 and 81.0 months,P=0.005 and<0.001,respectively).Evaluation the MRI features of HCC showed a good intra-and inter-observer consistency(Kappa values were 0.803-0.961 and 0.772-0.923,respectively).Univariate and multivariate logistic regression analysis and collinearity analysis showed the independent risk factors of early recurrence were alpha fetoprotein(AFP)(OR=8.459,P=0.002),albumin(ALB)(OR=3.842,P=0.016),multiple tumors(OR=3.842,P=0.008)and peritumoral enhancement(OR=4.127,P=0.023),which were further incorporated into predictive nomograms.Internal validation results demonstrated the area under the curve(AUC),sensitivity,specificity,and accuracy of the receiver operating characteristic(ROC)curve were 0.839,68.1%,93.1%,83.2%,respectively.The calibration curve illustrated the predicted curve and bias-corrected curve of nomogram was close to the ideal curve.Hosmer-Lemeshow test showed there was no significant difference between the predicted results of nomogram and the actual results(P=0.424).Conclusions An interval of 12 months after RFA was the optimal cut-off value for defining early recurrence and late recurrence.The nomogram,integrated by clinical and radiographic features,could potentially predict early recurrence of HCC after RFA and may offer useful guidance for individual treatment or follow up.Purpose To analysis the clinical,radiographic features and radiomic features based on MRI for late recurrence after radiofrequency ablation of HCC,to bulid the clinical-radiological model,radiomics model,combined model and to compare the performance for predicting RFS of late recurrence by these models.Materials and methods 150 patients with HCC after RFA follewed by late recurrence or no recurrence in Cancer Hospital,Chinese Academy of Medical Sciences from January 2012 to December 2017 were identified.Regions of interest(ROI)were manually delineated and radiomic features were extracted on pretreatment T2WI and dynamic enhanced-MRI including arterial,portal venous,delayed phase images.Intraclass correlation coefficient(ICC),the least absolute shrinkage and selection operator algorithm(LASSO)-Cox and 10-fold were used to select the optimized features and then built a radiomics signature.Patients were stratified into low-and high-risk group by radiomic score.Kaplan-Meier survival curve and log-rank test were used to compare the RFS of the two groups.Risk factors in clinical,radiographic features and Radscore for late recurrence were identified by Cox regression analysis.Clinical-radiological model,radiomics model,combined model was bulit by the independent risk factor and compared by concordance index(C-index)and integrated discrimination improvement(IDI).The nomogram was developed by the optimal model and validated.Results The radiomics signature is comprised by 11 selected features,which were significant associated with late recurrence.In the training and validation set,the RFS of low-and high-risk group stratified by radiomic score were statistically different(both P<0.001).Cirrhosis(HR=2.450,P=0.033),tumor margin(HR=2.494,P=0.004),and Radscore(HR=4.407,P<0.001)were independent risk factors for late recurrence.The C-index of the combined model was 0.794 and 0.771 in the training and validation set,respectively,which was significantly higher than that of the clinical-radiological model(training set:C-index=0.650,IDI=0159,P=0.007;validation set:C-index=0.675,IDI=0.141,P<0.001),but not significantly higher than that of the radiomics model(training set:C-index=0.675,IDI=0.048,P=0.120;validation set:c-index=0.746,IDI=0.048,P=0.060).The calibration curve showed the predicted results of nomogram deveploed by combined model was close to the actual situation.Conclusion The radiomics signature could be used as prognostic biomarker to predict late recurrence of HCC after RFA.With the addition of radiomics signature,the combined model can significantly improve the predictive efficiency of late recurrence compared to clinical-radiological model,which help clinicians to individualize the treatment of HCC patients.