AI Predicts Cancer in Routine Scans

Mayo Clinic’s AI spots pancreatic cancer up to three years early in routine scans, potentially saving lives before tumors even appear.

Story Highlights

REDMOD AI analyzes CT scans automatically, detecting 73% of cases at a median 16 months before diagnosis.
Outperforms specialists by nearly double overall and three times in scans over two years prior.
Validated on 2,000 scans from multiple institutions, mirroring real-world workflows.
Targets high-risk patients like those with new-onset diabetes, advancing to AI-PACED trials.
Part of Mayo’s Precure initiative to boost survival from 13% to 44% if caught early.

REDMOD AI Breakthrough Details

Mayo Clinic researchers developed REDMOD, a radiomics-based model that examines tissue textures in routine abdominal CT scans. The AI identifies subtle cancer signatures in pancreases appearing normal. Validated on nearly 2,000 prediagnostic scans from diverse institutions, REDMOD detected 73% of cases at a median 16 months before clinical diagnosis. It nearly doubled specialists’ detection rate and tripled it for scans over two years prior. Published in Gut, the study proves reliability across imaging protocols.

Lead Researchers and Mayo Clinic’s Role

Ajit Goenka, M.D., Mayo Clinic radiologist and senior author, leads REDMOD development. Goenka highlights the AI’s ability to capture faint changes experts miss in normal-appearing tissue. Mayo’s Pancreatic Cancer Early Detection Research Program drives this via its Radiology, AI, and Informatics departments. The Precure initiative integrates REDMOD for pre-symptomatic detection. Collaborations provide multi-site data, ensuring broad applicability without manual segmentation delays.

Overcoming Pancreatic Cancer’s Detection Barriers

Pancreatic cancer kills quickly because 85% spread before diagnosis, yielding 13% five-year survival. Localized cases reach 44% survival with surgery. Tumors evade early sight, but REDMOD spots precursors in everyday CTs done for other issues. High-risk groups like new-onset diabetes patients benefit most, paired with tools like END-PAC score. This evolution from Mayo’s 2025 AI, which flagged signals a year early, shows model progress grounded in peer-reviewed gains.

AI processes scans in seconds versus 20-30 minutes for experts, scaling to underserved areas.

Next Steps in AI-PACED Trials

REDMOD advances to the AI-PACED prospective study, enrolling high-risk patients to test real-world integration. Trials assess false positives, serial scan stability, and outcomes. Mayo announced validation in April 2026, building on February 2025 precursors. Goenka states the AI reliably flags invisible signatures across settings. Success could flag risks during routine visits, enabling curative intervention before symptoms.

Potential Impacts on Survival and Care

Short-term, AI-PACED refines flagging in diabetes patients and others via routine scans. Long-term, earlier detection cuts late-stage costs and lifts survival toward 44%. Radiologists gain doubled aid; patients get hope against a projected No. 2 cancer killer by 2030. Scalable AI democratizes access, aligning with values of efficient, innovative medicine that prioritizes lives over bureaucracy.