AI Model Redmod Detects Pancreatic Cancer 475 Days Before Clinical Diagnosis

AI spots cancer long before human doctors

Scientists at Mayo Clinic and collaborating institutions have developed an artificial intelligence model named Redmod that can detect subtle abnormalities on standard CT scans an average of 475 days before a patient receives a pancreatic cancer diagnosis. The findings were published in a new study.

Pancreatic cancer ranks among the deadliest forms of cancer. Tumors of this type rarely produce symptoms in early stages and often remain invisible on medical imaging until the disease has progressed to an advanced stage. According to the study, more than 85% of cases are detected at a point when treatment is limited to symptom management.

Why this matters

Early detection of pancreatic cancer fundamentally changes a patient's outlook. The researchers emphasized that identifying the tumor at a pre-clinical stage opens the door to surgical intervention and other effective treatments that are unavailable when the disease is caught late.

The study authors highlighted the significance of the detection window, noting that early identification could substantially improve cure rates and overall survival. Their modeling shows that increasing the proportion of localized pancreatic ductal adenocarcinomas from 10% to 50% would more than double survival rates. Timing of diagnosis, the experts stated, is the single most important factor determining treatment outcomes.

How Redmod works and its accuracy

Redmod analyzes patterns in CT images that are imperceptible to the human eye. The model was trained and tested on scans from over 1,400 individuals. Among them were 219 patients whose earlier scans had been assessed as normal by radiologists but who subsequently developed pancreatic cancer.

In a head-to-head comparison, the AI significantly outperformed radiologists: Redmod correctly identified 73% of cases versus 39% for human specialists. For scans taken more than two years before diagnosis, the AI's advantage grew even wider — 68% compared to just 23% for radiologists.

The model showed consistent performance across different hospitals and scanner types. It correctly classified more than 80% of scans from individuals who did not develop cancer, indicating an acceptable level of specificity.

Outlook and limitations

The researchers noted that Redmod could be used to flag high-risk patients for closer monitoring. However, before the tool can be integrated into routine clinical practice, prospective trials — real-world screening studies — will be necessary to validate its effectiveness.

Should those trials confirm the model's capabilities, the technology could trigger a paradigm shift in how one of the most aggressive cancers is diagnosed. Previously, in October 2025, Google in collaboration with Yale University unveiled a foundation model with 27 billion parameters designed to understand the "language" of individual cells, reflecting a broader trend of AI application in biomedicine.