Quantitative MRI Markers Detect Postpancreatitis Changes and Diabetes: A Prospective Cohort Study.

INTRODUCTION

Pediatric acute pancreatitis (AP) prevalence is increasing, yet early imaging markers identifying progression to acute recurrent pancreatitis (ARP) and pancreatitis-related endocrine dysfunction (DM/Pre-DM) are unknown. Quantitative magnetic resonance imaging (MRI) measures including T1 relaxation, T1 signal intensity ratio, proton density fat fraction (PDFF), and pancreatic volume may help characterize disease stage but have not been prospectively evaluated after a single AP episode.

METHODS

This prospective, single-center study enrolled participants aged 5-30 years. Participants were imaged ≥4 weeks after their most recent AP episode and classified as single AP (n = 21), ARP (n = 23), or DM/Pre-DM (n = 16). 3T MRI included Modified Look-Locker Inversion Recovery (MOLLI)-based T1 mapping, mDixon Quant PDFF, and pancreatic volumetric assessments. Clinical, laboratory, and genetic data were collected. Group comparisons, correlation analyses, analysis of variance, and receiver operating characteristic analyses with repeated 10-fold cross-validation were performed.

RESULTS

Sixty participants (median age 15.6 years) completed imaging. ARP showed higher T1 relaxation (1014 vs 819 ms, P = 0.015) and pancreatic PDFF (3.7% vs 1.7%, P = 0.021) than AP. DM/Pre-DM participants had the highest liver PDFF (9.9 vs 3.6% in AP and 2.7% in ARP, P = 0.024). Pancreatic volume, adjusted for body surface area, did not differ significantly. Liver PDFF was the strongest single predictor of DM/Pre-DM (area under the curve 0.77). A combined model including liver PDFF, pancreatic PDFF, and pancreatic volume improved discrimination (area under the curve 0.82).

DISCUSSION: Quantitative MRI demonstrates distinct imaging features in children and young adults with ARP and Pre-DM/DM. Increased T1 relaxation and pancreatic fat characterize ARP, whereas hepatic fat is most strongly associated with endocrine dysfunction. Multimarker models improve classification and require validation in larger cohorts.