Prediction of nonalcoholic fatty liver disease (NAFLD) activity score (NAS) with multiparametric hepatic magnetic resonance imaging and elastography

Objectives To investigate the use of MR elastography (MRE)-derived mechanical properties (shear stiffness (|G*|) and loss modulus (G '')) and MRI-derived fat fraction (FF) to predict the nonalcoholic fatty liver disease (NAFLD) activity score (NAS) in a NAFLD mouse model. Methods Eighty-nine male mice were studied, including 64 training and 25 independent testing animals. An MRI/MRE exam and histologic evaluation were performed. Pairwise, nonparametric comparisons and multivariate analyses were used to evaluate the relationships between the three imaging parameters (FF, |G*|, and G '') and histologic features. A virtual NAS score (vNAS) was generated by combining three imaging parameters with an ordinal logistic model (OLM) and a generalized linear model (GLM). The prediction accuracy was evaluated by ROC analyses. Results The combination of FF, |G*|, and G '' predicted NAS > 1 with excellent accuracy in both training and testing sets (AUROC > 0.84). OLM and GLM predictive models misclassified 3/54 and 6/54 mice in the training, and 1/25 and 1/25 in the testing cohort respectively, in distinguishing between not-NASH and definite-NASH. Borderline-NASH prediction was poorer in the training set, and no borderline-NASH mice were available in the testing set. Conclusion This preliminary study shows that multiparametric MRI/MRE can be used to accurately predict the NAS score in a NAFLD animal model, representing a promising alternative to liver biopsy for assessing NASH severity and treatment response.