Overview
- A peer-reviewed study in Gut reports an AI called REDMOD found pre-clinical signals on routine CT about 475 days before diagnosis on average.
- The model identified 73% of future pancreatic ductal adenocarcinoma cases versus 39% for expert radiologists reading the same scans.
- Researchers trained and tested it on 219 pre-diagnostic cases and 1,243 matched controls, and external datasets showed good specificity around the low- to high-80s.
- The system uses radiomics to read subtle texture patterns and an automated step to outline the pancreas so results do not hinge on manual tracing.
- The team calls for prospective trials in high-risk and more diverse groups before clinical use, noting that earlier detection could move more patients into curative surgery and improve survival.