AI Model Spots Pancreatic Cancer 475 Days Before Doctors Can

A Breakthrough in Early Detection

An artificial intelligence framework can identify the earliest signs of pancreatic cancer on routine CT scans more than a year before doctors would typically catch it, according to a study published in the journal Gut. The tool, called REDMOD (Radiomics-based Early Detection MODel), was developed by researchers at Mayo Clinic and collaborators. It detected what the researchers describe as the 'invisible' signature of preclinical pancreatic ductal adenocarcinoma an average of 475 days before clinical diagnosis, picking up subtle tissue texture changes that the human eye and standard imaging cannot see.

Outperforming Radiologists

REDMOD proved far more capable than experienced radiologists at spotting early malignant changes. The model was nearly twice as sensitive as radiologists, at 73% compared with 39%, in accurately identifying preclinical disease. For cases detected more than two years before clinical diagnosis, the gap widened further: REDMOD was nearly three times as accurate, at 68% versus 23%. The model also correctly classified more than 81% of scans in an independent group of 539 patients across several hospitals as cancer-free, and 87.5% in the public NIH-PCT dataset of 80 patients. When the same patient was scanned again months earlier, REDMOD delivered consistent results 90 to 92% of the time.

A Disease Where Time Is Everything

Pancreatic cancer remains the deadliest major cancer in the United States, with a five-year survival rate stuck at 13%. It is the third-leading cause of cancer-related deaths, with an estimated 67,530 Americans expected to be diagnosed and 52,740 to die from the disease this year. However, when detected at a localised stage before it spreads, the five-year survival rate jumps to 44%. That stark difference is what makes REDMOD's early detection window so consequential.

Road to the Clinic

The framework includes automated pancreatic segmentation, eliminating the need for radiologists to manually outline the borders of the pancreas on each scan. The researchers cautioned that prospective validation in high-risk patients, such as those with unexplained weight loss and newly diagnosed diabetes, is needed before the tool can be widely adopted. They described REDMOD as a significant advance towards shifting pancreatic cancer detection from a late-stage symptomatic diagnosis to proactive pre-clinical interception.