AI Model Detects Early Pancreatic Cancer with 92% Accuracy from C...

Researchers at Johns Hopkins University and the UK National Health Service published results of a clinical validation study showing their AI model Prism achieved 92% sensitivity in detecting stage one and two pancreatic cancer from routine abdominal CT scans — a disease that typically presents no symptoms until stage three or four when survival rates are below 15%. The model was trained on 24,000 annotated CT scans and validated on a separate dataset of 9,000 scans from 18 hospitals across five countries. Prism reduces radiologist reading time per scan by 60% and identifies micro-lesions as small as four millimetres that human radiologists routinely miss. Apollo Hospitals and Fortis Healthcare in India have signed agreements to integrate Prism into their radiology workflows starting in 2026. WHO estimates that early detection at stage one through tools like Prism could raise five-year pancreatic cancer survival from 11% to over 55%.