MR elastography as a biomarker for prediction of early and late recurrence in HBV-related hepatocellular carcinoma patients before hepatectomy

Purpose: To investigate the diagnostic performance of preoperative MR elastography (MRE) in predicting early recurrence (ER) and late recurrence (LR) of HCC after hepatectomy. Method: In total, 180 patients (median age, 52 years; interquartile range, 41-50 years; 161 men) who underwent conventional MRI and MRE before hepatectomy between December 2014 and April 2020 were retrospectively recruited. A preoperative clinic-radiologic model and a combined postoperative clinic-pathologic and radiologic model were built using quantitatively MRE-derived stiffnesses, and image features to predict tumor ER and LR after hepatectomy. The Cox proportional hazards model and ROC analyses were used to identify the value of parameters to predict ER and LR. Results: Seventy-three (40.5%) and 16 (8.9%) developed ER and LR after hepatectomy, respectively. For pre-diction of ER, the preoperative model integrated higher tumor stiffness (TS) (hazard ratio [HR],1.142; p < 0.001) with AFP >= 400 ng/mL (HR,1.761; p = 0.022), multifocal tumors (HR,3.229; p < 0.001) and lower ADC (HR,0.998; p = 0.017) variables; and the postoperative model incorporated higher TS, microvascular invasion, multifocal tumors, Child-Pugh class and ADC predictors. The two models provided comparable predictive per-formance (pre-0.812 vs. post-0.834, p = 0.283). Moreover, TS alone had a high sensitivity (90.4%) for pre-dicting ER. Liver stiffness (LS) (HR, 1.757; p < 0.001) was the only independent predictor for LR in multivariate analysis in both the pre-and postoperative models with high specificity (90.0%), and its AUC with an optimal cut-off of 3.62 kPa was 0.860. Conclusions: Quantitative MRE-based stiffness is a useful biomarker for preoperative prediction of ER and LR of HCC.

Automated summary

This study demonstrates that MR elastography is a useful method for predicting early recurrence and late recurrence of HCC. Both preoperative and postoperative models showed good predictive performance.