Overview
- The study, which appeared Tuesday in Gut, found the REDMOD system identified preclinical pancreatic ductal adenocarcinoma a median 16 months before diagnosis with about 88% specificity and 73% sensitivity.
- In head-to-head reads, the model nearly doubled specialists’ detection on the same scans that had been reported as normal, and for images more than two years before diagnosis it reached 68% versus 23% for radiologists.
- Researchers validated the fully automated tool on nearly 2,000 CT scans from multiple hospitals and scanners in workflows that mirror clinical practice, and its risk scores stayed consistent on repeat scans months apart.
- The team has moved into prospective testing through the AI-PACED study to see how the tool fits into care, measure false positives, and assess performance in more diverse patient groups.
- Earlier detection could move more patients into a window when surgery is possible, and modeling suggests lifting localized diagnoses from 10% to 50% could more than double survival for this lethal cancer.