AI Model Detects Pancreatic Cancer Up to 3 Years Earlier Than Doctors in Early Test

The Radiomics-based Early Detection Model (REDMOD), described in a study published April 28, 2026 in the journal Gut, identified 73% of early-stage pancreatic cancer cases from 2,000 existing CT scans previously signed off as normal. The scans were taken on average 16 months before diagnosis. REDMOD's sensitivity gain over radiologists was nearly twofold overall and almost threefold when detecting cancer more than two years prior to diagnosis.

Pancreatic cancer detection challenges

• ▪The five-year survival rate for pancreatic cancer in the U.S. is about 12% to 13%.
• ▪The process of pancreatic cancer development starts 10 to 15 years before diagnosis.
• ▪The early stages of pancreatic cancer often don't trigger any symptoms, so the disease is often advanced at the point of diagnosis.
• ▪By the time pancreatic cancer tumors are visible via tissue sampling and imaging tests including CT scans, the cancer is often terminal.

REDMOD AI model development

• ▪The Radiomics-based Early Detection Model (REDMOD) converts CT scan images into mathematical representations to identify early signs of pancreatic cancer.
• ▪REDMOD segments the pancreas organ and builds a 3D model from 2D CT scan images.
• ▪REDMOD evaluates the pancreas structure pixel by pixel, quantifying the degree to which each pixel differs from the rest of the organ.
• ▪REDMOD was described in a study published April 28, 2026 in the journal Gut.
• ▪Dr. Ajit Goenka is a radiologist and nuclear medicine specialist at the Mayo Clinic in Rochester, Minnesota.

Early detection performance results

• ▪REDMOD successfully identified 73% of early-stage pancreatic cancer cases in the test sample.
• ▪About one-seventh of the 2,000 CT scans tested belonged to patients who later went on to develop pancreatic cancer.
• ▪The CT scans analyzed by REDMOD had been taken on average 16 months before the patient's actual pancreatic cancer diagnosis.
• ▪REDMOD was tested on a sample of 2,000 existing CT scans that had all been previously signed off as normal.

False positive rates comparison

• ▪Human radiologists correctly identified disease-free patients an average of 92.2% of the time.
• ▪REDMOD correctly identified disease-free patients 81.1% of the time.
• ▪Human radiologists were less likely than REDMOD to flag a healthy patient incorrectly.

Clinical implementation timeline

• ▪Dr. Ajit Goenka hopes REDMOD could be routinely implemented in the clinic within the next five years.
• ▪The REDMOD development team is currently running clinical trials to further validate that the detection strategy works in practice.

Complementary diagnostic approaches

• ▪Tatjana Crnogorac-Jurcevic is developing urine-based tests for early pancreatic cancer detection.
• ▪Tatjana Crnogorac-Jurcevic is a professor of molecular pathology and biomarkers at Queen Mary University of London.

1 source