Pancreatic cancer: AI spots signs on CT scans an average of 475 days before diagnosis

Artificial intelligence can detect signs of pancreatic cancer on computed tomography scans long before doctors see the tumour, according to a new study. The model, REDMOD, analyses fine tissue-texture patterns in abdominal CT scans and automatically outlines the pancreas. Researchers led by Sovanlal Mukherjee and Ajit Harishkumar Goenka published the results online on Oct. 26, 2026, in the BMJ Group journal Gut, ScienceDaily reported.

The team examined CT scans of 219 patients from several hospitals in whom radiologists saw no cancer at the time, but who were later diagnosed with pancreatic ductal adenocarcinoma. Scans from 1,243 people who stayed cancer-free for up to three years served as a comparison. REDMOD flagged the changes an average of 475 days – about 1.3 years – before clinical diagnosis. Its sensitivity was 73%, compared with 39% for experienced radiologists; for scans taken more than two years before diagnosis, it was 68% versus 23%.

In an independent group of 539 patients, the model correctly classified just over 81% of scans as cancer-free, and 87.5% in the public dataset of the U.S. National Institutes of Health. The authors stress that more research is needed before the method can enter routine care – above all prospective studies in high-risk groups, such as people with unexplained weight loss or newly diagnosed diabetes. They also said the study population was not ethnically diverse enough. The release makes no mention of regulatory approval.