Diffusion-weighted MRI-derived μDiff index and extracellular volume fraction for severity classification of acute pancreatitis

Objectives To evaluate the diffusion-weighted imaging (DWI)-derived quantitative index (μ Diff ) and T1 mapping-derived extracellular volume (ECV) fraction for distinguishing mild acute pancreatitis (AP) from the combined moderately severe or severe category, and to compare their discrimination with magnetic resonance severity index (MRSI) and clinical scores. Materials and methods This retrospective study enrolled patients with AP who underwent abdominal MRI early after admission between 2024 and 2025. MRI protocols included conventional imaging, DWI, and pre- and postcontrast T1 mapping. The DWI-derived μ Diff index was calculated from signal intensities at b values of 200 and 1500 s/mm 2 , whereas ECV was derived from pancreatic and blood-pool T1 values. AP severity was classified according to the 2012 revised Atlanta classification. Five single-predictor and four two-predictor logistic models were evaluated using AUCs with 95% confidence intervals, paired DeLong comparisons, and 2,000-resample bootstrap internal validation; calibration and exploratory decision-curve analyses were also performed. Results Among 140 patients with AP, 99 had mild AP and 41 had moderately severe or severe AP. Both μ Diff and ECV increased with severity. μ Diff showed an AUC of 0.896 (95% CI , 0.836–0.956), with no statistically significant difference from MRSI (0.878; 95% CI , 0.811–0.946; DeLong p = 0.720). The μ Diff + MRSI model achieved an apparent AUC of 0.972 (95% CI , 0.948–0.996) and an optimism-corrected AUC of 0.969, with improved discrimination over μ Diff and MRSI alone (Holm-adjusted p = 0.019 and p = 0.003, respectively). Its corrected calibration slope and Brier score were 0.911 and 0.067. Conclusions The DWI-derived μ Diff index showed good discrimination of Atlanta-defined AP severity, with performance statistically comparable to MRSI. Their combination improved discrimination and retained high performance after internal validation, supporting their complementary value for severity classification.