Non-alcoholic Fatty Liver Disease (NAFLD/Metabolic dysfunction-associated fatty liver disease, MAFLD) [1], affecting about 25% globally [2], ranged from simple steatosis (>5% fat) to Nonalcoholic Steatohepatitis (NASH) [3]. NASH, defined by steatosis, inflammation, ballooning, and fibrosis [4], was a critical precursor to cirrhosis and cancer. Early diagnosis was crucial for effective intervention [5]. While liver biopsy remained the diagnostic gold standard [6,7], its invasiveness and limitations underscored the urgent need for non-invasive, imaging-based biomarkers to stratify NAFLD severity and detect NASH.

Habitat imaging emerged as a novel radiological approach for quantifying spatial heterogeneity within tissues. This technique partitioned a volume of interest (VOI) into subregions (‘habitats’) with distinct imaging signatures, enabling precise characterization of tissue microenvironments through computation of volume fraction for each habitat [[8], [9], [10]]. Currently, habitat analysis has demonstrated significant prognostic and diagnostic value in oncology, including applications in hepatocellular carcinoma [11], breast cancer [12], and glioma [13]. However, its potential for evaluating diffuse hepatic pathologies such as NAFLD remained unexplored.

Recent advances in multi-b-value diffusion-weighted imaging (DWI) have facilitated the use of quantitative models, including mono-exponential DWI, IVIM, SEM, and DKI, to probe perfusion, diffusion heterogeneity, and microstructural complexity in NAFLD [[14], [15], [16], [17]]. While these models provided valuable scalar metrics, they typically overlooked the spatial heterogeneity inherent to NAFLD pathophysiology.

Building upon our previous work [16,17] with multiparametric DWI metrics from the same dataset, we hypothesized that integrating these parameters into a habitat imaging framework could better capture the spatial heterogeneity of liver pathology, an aspect not addressed by conventional region-of-interest averaging as used in our prior study. This re-analysis proposed an alternative method to summarize diffusion information, with the aim of characterizing spatially resolved microenvironmental patterns. Consequently, the objective of this methodological study was to develop and validate a DWI-based habitat imaging framework for the non-invasive stratification of NAFLD in a diet-induced rabbit model, using histopathology as the reference standard. Its primary focus was on assessing methodological feasibility, rather than on demonstrating superior diagnostic performance relative to our previous study.