AI model finds pancreatic cancer over a year early

The new AI system REDMOD can spot pancreatic cancer more than a year before it shows up on conventional scans, picking up subtle tissue changes an average of 475 days ahead. Catching the disease at stage 0 could mean earlier treatment and better odds of survival for this deadly cancer.

The study, published in Gut, shows REDMOD outperforms experienced radiologists, reaching a sensitivity of 73% compared with 39%. Researchers analyzed CT scans from 219 patients who were later diagnosed with pancreatic ductal adenocarcinoma, along with scans taken 3 to 12 months, 12 to 24 months, or more than 24 months before diagnosis. A comparison group of 1,243 cancer-free people was matched for age, sex and scan dates. REDMOD identified early warning signs an average of 475 days before diagnosis, with consistent results across repeated scans (90–92% agreement).

Limitations and future steps

The authors caution that the system still needs prospective testing, especially in high-risk groups such as people with unexplained weight loss or recently diagnosed diabetes. The study population lacked ethnic diversity, which could affect how well the system works for different groups. Until it is validated in routine practice, REDMOD remains a promising research tool rather than a clinically ready one.

If validated, REDMOD could shift pancreatic cancer diagnosis from a late-stage event to a pre-clinical intervention. Modelling studies cited by the researchers suggest that finding more cancers at a localized stage could more than double survival rates, offering real hope for patients who would otherwise face a late diagnosis.