AI‐Powered 3D Liver Fat Volumetric Map for Clinical‐Grade Steatosis Assessment From Cross‐Modality MRI

Magnetic resonance imaging‐proton density fat fraction (MRI‐PDFF) is widely used for noninvasive assessment of metabolic dysfunction‐associated steatotic liver disease, yet its dependence on manual analysis and the limited anatomical context of PDFF maps constrain its reproducibility and diagnostic utility. Here, using 129 abdominal MRI datasets from 117 participants acquired with separate 3D breath‐hold T1‐weighted dual‐echo and PDFF sequences within the same examination session, we present an automated liver fat quantification (ALFQ) system that integrates a pseudo‐3D U‐Net trained on structurally informative T1‐weighted dual‐echo images with a cross‐modality alignment framework to overcome slice misalignment and vascular interference. ALFQ generates a 3D liver fat volumetric map (LFVM) and shows concordance with manually derived liver fat fraction measurements (Pearson r = 1.00; ICC = 0.99), while providing steatosis grading (F1 score = 0.93) and shortening analysis time by nearly 83% relative to manual workflows, thereby improving efficiency. Moreover, the LFVM delivers a voxel‐level characterization of steatosis and a robust quantitative measure with increased information entropy, enabling sensitive assessment of therapeutic response over time. These features establish LFVM as an effective and precise diagnostic measure and position ALFQ as a clinically robust framework for anatomically informed liver fat map, advancing AI‐enabled precision hepatology.