Assessment and integration of multiparametric MRI for liver fibrosis staging in rat non-alcoholic steatohepatitis: Evaluation of diagnostic efficiency and model interpretation.

OBJECTIVES

Multiparametric magnetic resonance imaging (mpMRI) techniques, including intravoxel incoherence motion (IVIM), iterative decomposition of water and fat with echo asymmetry and least-squares estimation quantification sequence (IDEAL IQ), T2* mapping and T2 mapping, were employed to develop and validate a predictive model for non-alcoholic steatohepatitis (NASH) diagnosis and liver fibrosis (LF) staging in rats. The combined model was interpreted using SHapley Additive exPlanations (SHAP) values for model interpretation.

MATERIALS AND METHODS

160 healthy Sprague-Dawley (SD) rats were divided into control (n = 24) and experimental (n = 136) groups, and the 12-week and 16-week groups were injected intraperitoneally with carbon tetrachloride (CCl4) for 4 weeks, one month before the final feeding period. All rats were subjected to pathological examination to determine LF stage. Upon the study's completion, 147 SD rats were assessed for liver fibrosis.

RESULTS

84 SD rats were diagnosed with NASH and 31, 10, and 43 rats were histologically diagnosed with no fibrosis (F0), early LF (F1-F2), and advanced LF (F3-F4). For diagnosis of NASH and staging of liver fibrosis associated with NASH, a combined mpMRI prediction model has a higher area under the receiver operating characteristic(ROC) curve (AUC) than uniparameters, especially in advanced stages of fibrosis, with an AUC of 0.929 for the combined model. In SHAP, the fat fraction(FF) value contributes most to the model for diagnosing NASH and advanced liver fibrosis, while the T2 value contributes most for diagnosing liver fibrosis and the apparent diffusion coefficient (ADC) value contributes most for diagnosing liver cirrhosis.

CONCLUSIONS

The mpMRI could be used to evaluate the severity of liver fibrosis in the context of NASH. Combined with SHAP value analysis, this approach can help to understand the contribution of each mpMRI feature to the predictive model.