APEX-NET: AI model for pancreatitis diagnosis and severity prediction on non-contrast CT

APEX-NET, a deep learning model using non-contrast CT, achieved AUCs of 0.949-0.981 for acute pancreatitis diagnosis and 0.873 for severity prediction, matching senior radiologists in a multi-center study of 3383 patients.

• Retrospective and prospective study across 5 centers, 3383 patients (AP and non-AP).
• For severity prediction, macro-average AUCs were 0.873 (validation) and 0.872 (internal testing), outperforming NCCT-only model (p<0.05).
• In a reader study, APEX-NET performed comparably to senior radiologists and better than junior radiologists (p<0.05).

RadPigeon summaries are original and for information only. They are not clinical advice.