Overview
- Mayo Clinic researchers reported in Gut that their REDMOD system spotted pancreatic cancer risk on routine CT scans a median 475 days before diagnosis in a multi-hospital set of nearly 1,500 scans, in a disease with about a 13% five-year survival rate.
- The AI reached 73% sensitivity versus 39% for radiologists reading the same prediagnosis scans, and it was close to three times more accurate on images taken more than two years before diagnosis.
- The tool has moved into clinical trials to measure real world accuracy, false positives and performance across diverse patients before any broad rollout.
- Study leaders say hospitals are not yet set up for AI-only alerts, raising concrete questions about who receives a flag, how to escalate a normal-looking scan, how to explain risk to symptom-free patients and how payers cover follow-up testing.
- Health care and industry groups are advancing related efforts, including a UPMC test that detected nearly twice as many bile duct cancers as standard pathology and a Bristol Myers Squibb–Microsoft project using FDA-cleared imaging algorithms to spot early lung disease.