AI spots invisible pancreatic cancer clues years before diagnosis
An AI system named REDMOD analyzes routine abdominal CT scans to reveal ultra-early, radiomics-based tissue changes linked to pancreatic ductal adenocarcinoma, identifying cancer on average 475 days before diagnosis and outperforming expert radiologists in sensitivity (73% vs 39%). It includes automated pancreas segmentation and was tested across multiple cohorts, achieving ~81% cancer-free accuracy and 87.5% in NIH-PCT data; similar results were reproduced on repeated scans in 90–92% of cases. While promising, it requires prospective validation and testing in high-risk groups before clinical deployment, and time-to-diagnosis improvements could shift detection toward stage 0 and improve survival.
