CT-based Multimodal Model for Early Severity Prediction in Hyperlipidemic Acute Pancreatitis: A Multicenter Study.

RATIONALE AND

OBJECTIVES

To develop and validate a multimodal model integrating pancreatic parenchymal and peripancreatic fat radiomics with independent clinical predictors for early severity stratification in hyperlipidemic acute pancreatitis (HLAP).

MATERIALS AND

METHODS

This retrospective multicenter study included 356 patients with HLAP from three centers. HLAP was defined as acute pancreatitis (AP) with serum triglycerides ≥11.3 mmol/L or 5.65 to 11.3 mmol/L, accompanied by lactescent serum after exclusion of other etiologies. Patients from Center 1 were assigned to the training (n = 193) and internal validation (n = 83) cohorts, whereas patients from Centers 2 and 3 formed the external test cohort (n = 80). One noncontrast abdominal computed tomography examination acquired within 72 hours of symptom onset was analyzed per patient. Clinical, radiomics, and fusion models were developed using four machine-learning algorithms. Performance was evaluated using the area under the receiver operating characteristic curve (AUC), calibration curves, decision curve analysis, and DeLong tests.

RESULTS

The cohort included 107 patients with severe AP and 249 with nonsevere AP. The median age was 37.0 years (interquartile range, 31.75-45.0), and 268 patients (75.3%) were male. The final pancreatic parenchyma-peripancreatic fat-clinical combined model achieved AUCs of 0.947 (95% confidence interval [CI], 0.897-0.986) and 0.846 (95% CI, 0.735-0.940) in the internal validation and external test cohorts, respectively, with external sensitivity and specificity of 0.611 and 0.871. DeLong testing showed significant improvements over the pancreatic parenchyma radiomics, 15 mm peripancreatic fat radiomics, and clinical models in internal validation (all p < 0.05), but not in external testing (all p > 0.05).

CONCLUSION

The proposed multimodal model showed good discrimination and interpretability for early HLAP severity stratification and may support individualized clinical decision-making.