REDMOD AI Identifies Pancreatic Cancer Signals on CT Before Diagnosis

Researchers evaluated REDMOD, an automated AI system that segments the pancreas and analyzes CT texture patterns, using scans from 219 patients whose scans had initially been read as showing no pancreatic cancer. All were subsequently diagnosed with pancreatic ductal adenocarcinoma. The scans were compared with those from 1,243 people who remained cancer-free for up to three years; the groups were matched by age, sex and scan date. The study also assessed cancer-free classification in an independent 539-patient group and a separate 80-person public dataset.

REDMOD identified imaging signatures an average of 475 days before clinical diagnosis. Its sensitivity was 73%, compared with 39% for experienced radiologists. For scans taken more than two years before diagnosis, sensitivity was 68% for REDMOD and 23% for radiologists. The system correctly classified just over 81% of scans as cancer-free in the independent group and 87.5% in the public dataset.

The findings suggest that AI analysis may detect changes not recognized on routine CT review, but they do not establish that using REDMOD improves diagnosis, treatment or survival. The study was retrospective, the patient population lacked sufficient ethnic diversity, and the authors call for prospective validation, particularly among people at increased risk, before routine use can be considered.