Pancreatitis or inflammation of the pancreas comes in both acute and chronic forms and has long been appreciated as a risk factor for developing pancreatic cancer. While common and in most cases mild, severe acute pancreatitis can be fatal due to severe inflammation causing local damage that can have serious systemic consequences. The ability to predict which patients who initially present with acute pancreatitis will develop severe disease is currently limited. With the adoption of AI, new models may help to more accurately predict patients who will develop severe and life-threatening disease, and in the future may be able to help predict those that may develop pancreatic cancer.
Peter Hegyi, MD, PhD and Tamas Gonda, MD, were awarded a collaborative Hirshberg Foundation Seed Grant in 2023 to evaluate the use of AI and machine learning to predict which patients will develop severe acute pancreatitis and to predict those who may develop pancreatic cancer. Dr. Hegyi, Professor and Director of the Institute for Pancreatic Diseases at Semmelweis University in Budapest, Hungary, and Dr.Tamas Gonda, Associate Professor of Medicine in Gastroenterology at New York University Grossman School of Medicine, sought build a database of well characterized acute pancreatitis patients across NYU and Semmelweis University to develop a model to predict acute pancreatitis severity. They also set out to build a multi-institutional coalition to identify cases and imaging where pancreatic cancer was diagnosed with 2 years of acute pancreatitis to then extend this deep learning model to predict which acute pancreatitis patients may later develop pancreatic cancer.
The team developed and validated a deep learning model that predicts acute pancreatitis severity using contrast-enhanced CT scans. Their model was trained on over 10,000 CT studies and then fine-tuned on over 500 additional cases where severity outcome was known. In both internal and external data sets, the new deep-learning model outperformed established prediction tools. The model was then used to analyze a cohort of images from patients with known outcomes (also called a retrospective analysis) and it was able to identify 73% of patients that went on to develop severe acute pancreatitis. These results support the continued validation of early, automated risk triage for acute pancreatitis. A recent paper, Deep learning-based prediction of acute pancreatitis severity from abnormal CT with multicenter external validation, was published in the May 2026 journal Radiology Advances.
Current and future research will focus the new machine learning algorithms on images from patients with acute pancreatitis who later develop pancreatic cancer. This will train the programs to identify image-based changes in the hopes of generating a predictive model to stratify these patients’ cancer risk at the time of acute pancreatitis onset. The ability to use image-based biomarkers for this high-risk population will lead to earlier pancreatic cancer diagnoses and better treatment options.
Early detection remains one of the greatest opportunities to change outcomes for pancreatic cancer patients. The Hirshberg Foundation is committed to funding bold and innovative research that makes early detection possible. Better diagnostic approaches offer the greatest opportunity for successful treatment and improved survival, yet too many patients are still diagnosed at advanced stages. AI-driven imaging represents one of many promising approaches with the potential to change how pancreatic cancer is found and treated. Continued funding is critical to move these discoveries from the laboratory into real-world clinical care. The power to detect pancreatic cancer sooner helps expand treatment options and save lives.
