A CT-based radiomics model for classification of chronic pancreatitis: new biomarkers for diagnosis and severity staging.

OBJECTIVES

To develop and validate CT-based radiomics models for the identification of chronic pancreatitis (CP) and selected CP-related complications, and to explore associations between radiomics-derived features and clinical measures.

MATERIAL &

METHODS

This retrospective, multicenter study included a training cohort of 349 subjects (201 CP; 148 healthy controls) from Aalborg University Hospital. Test cohort comprised 109 subjects, including 41 pancreas-healthy controls from Aalborg and an external cohort of 68 CP patients from Bergen. Portal venous phase CT scans were automatically segmented, and radiomics features were extracted using PyRadiomics. AI models were trained to classify CP and identify CP-related complications, including exocrine pancreatic insufficiency (EPI), diabetes, and pain. Model performance was assessed using area under the receiver operating characteristic curve (AUC) with 95% confidence intervals (CI).

RESULTS

The CP classification model demonstrated high discriminative performance with an AUC of 0.97 (95% CI: 0.94-0.99). The EPI model showed moderate discriminative performance (AUC 0.80, 95% CI: 0.66-0.88). In contrast, the diabetes and pain models demonstrated lower discriminative performance, with AUCs of 0.63 (95% CI: 0.47-0.77) and 0.59 (95% CI: 0.37-0.67), respectively. Radiomics-derived probability scores correlated significantly with fecal elastase levels (p < 0.001) and increased with greater functional disease severity (p = 0.004).

CONCLUSION

CT-based radiomics can accurately classify CP and reflect exocrine functional impairment. However, performance for diabetes and pain was limited, and clinical utility beyond established clinical assessments remains to be demonstrated.