AI Detects Pancreatic Cancer Before It Becomes Visible on CT

Pancreatic cancer is usually diagnosed too late for curative surgery. More than 85% of patients present with advanced disease, when resection is no longer possible.

Researchers at Mayo Clinic have developed an AI model that detects imaging changes months before pancreatic cancer becomes visible on routine CT scans.

Looking beyond the tumor

The model, called REDMOD, analyzes radiomic features throughout the pancreas instead of searching for a visible mass.

In a validation study of nearly 2,000 CT scans, including 219 scans from patients later diagnosed with pancreatic ductal adenocarcinoma, REDMOD identified 73% of prediagnostic cancers. The median lead time was about 16 months before clinical diagnosis.

Compared with specialist review alone, the model detected roughly twice as many early cancers and performed even better on scans obtained more than two years before diagnosis.

Why surgeons should care

For pancreatic cancer, timing determines treatment options.

Earlier detection could allow patients to undergo additional imaging, biopsy, and staging while disease is still confined to the pancreas and potentially resectable. Survival remains dramatically higher when surgery is performed before the cancer has spread.

REDMOD was validated across multiple institutions and CT platforms, and because it analyzes routine CT scans, it could be integrated into existing imaging workflows without additional testing.

What comes next

The study shows that AI can identify patients at risk earlier than current practice. It does not yet show that earlier detection improves survival.

That question is now being evaluated in Mayo Clinic's prospective AI-PACED study, which will assess how the model performs in routine clinical care.

Why It Matters

For pancreatic surgeons, the most valuable AI may be the one that finds operable disease before anyone else can see it.

Source: Mayo Clinic News Network