Preoperative gadoxetic acid-enhanced MRI combined with microvascular invasion can effectively improve the predictive efficacy of early recurrence in hepatocellular carcinoma : a Two- center Study

Abstract Objectives To compare the effective of preoperative factors based on gadoxetic acid-enhanced magnetic resonance imaging (MRI) and pre-/postoperative factors in predicting early recurrence of hepatitis B virus(HBV)-related single hepatocellular carcinoma (HCC) after surgical resection. Methods This is a retrospective two-center study, from July 2013 to November 2018 consecutive patients underwent surgery for HCC with preoperative gadoxetate disodium (EOB)-enhanced MR were enrolled, to training and testing sets. In the training sets, multivariate Cox regression analyses of preoperative and postoperative high risk predictors of early HCC recurrence (EHR) were performed. Preoperative and pre-/postoperative prediction models were constructed. In the training sets and testing sets, using the time-dependent receiver operating characteristic(ROC) curves to evaluate the discrimination of the two prediction models, area under curves (AUC) were calculated. The diagnostic performance of each risk features and combine features were also calculated. Recurrence-free survival (RFS) were analysed by the Kaplan–Meier method with the log-rank test. Results Three MRI features were independently associated with EHR: (nonhypervascular hepatobiliary phase(HBP) hypointense nodule (NHHN) (Hazard Ratio(HR) = 3.700), non-smooth tumour margin (HR = 2.781), and hepatobiliary phase (HBP) peritumoural hypointensity (HR = 2.614). After adding postoperative histopathologic factors, presence of microvascular invasion (MVI) and one MRI features were independently associated with EHR: MVI (HR = 3.885), NHHN (HR = 4.320). In the training sets, AUCs for the two models were 0.821 (preoperative prediction models), 0.890 (pre-/postoperative prediction models). In the testing sets, AUCs for the two models were 0.748 (preoperative prediction models), 0.794 (pre-/postoperative prediction models). The diagnostic performance of each risk factor and combine factors were all lower than the two prediction models. Conclusions In this two-center study, the preoperative and pre-/postoperative prediction models demonstrate good performance for predicting EHR, the AUC of the two prediction model better than only the MVI risk factor. In the preoperative prediction model, NHHN, non-smooth tumour margin, and HBP peritumoural hypointensity can be used preoperatively to estimate the risk of early recurrence after resection of a single HCC. Such models might be useful to guide treatment strategies and enable a step forward for precise medicine.