Pancreatic cancer kills roughly 52,740 Americans a year because it is almost always found too late. A Mayo Clinic model called REDMOD suggests the warning signs may already sit in scans hospitals performed for other reasons.
Published in the journal Gut, the study found REDMOD identified subtle imaging signatures of pancreatic ductal adenocarcinoma an average of 475 days before clinical diagnosis, and nearly twice as often as experienced radiologists reviewing the same images.
Reading texture the eye cannot resolve
REDMOD stands for Radiomics-based Early Detection Model. It measures fine-grained patterns in pancreatic tissue texture that are not visible on conventional CT, and it automatically segments the pancreas so no specialist has to hand-trace the organ.
Researchers analyzed abdominal CT scans from 219 patients across multiple hospitals. Every scan looked cancer-free to radiologists at the time, and every patient was later diagnosed. They were compared against scans from 1,243 matched controls.
In an independent test set the model reached 73% sensitivity against 39% for radiologists. For scans taken more than two years before diagnosis, the gap widened to 68% versus 23%.
The specificity question nobody should skip
REDMOD’s specificity was 81.1%, meaning roughly one in five healthy patients would receive a false positive flag. That trade-off is the core operational question for any health system weighing deployment.
The workflow gap is institutional rather than technical. If the model flags an asymptomatic patient today, who orders follow-up, and what does that follow-up look like?
This was retrospective. A prospective trial called AI-PACED is underway, but results are years away, and no evidence yet shows an earlier flag improves survival. The best near-term fit may be opportunistic screening in high-risk groups such as people with new-onset diabetes.
