What if a medical scan could reveal warning signs of cancer long before a tumor becomes visible?
Researchers in the United States have developed an artificial intelligence system capable of identifying subtle changes associated with pancreatic cancer in CT scans taken months or even years before patients received their diagnosis.
The technology, called REDMOD, was developed by researchers from Mayo Clinic and the University of Texas MD Anderson Cancer Center. It analyzes patterns in pancreatic tissue that may be too subtle for doctors to recognize through conventional scan interpretation.
In a study published in the medical journal Gut, the system identified approximately 73% of pancreatic cancer cases in an independent test group, compared with about 39% detected by experienced radiologists reviewing the scans without AI assistance.
The warning signs appeared a median of approximately 16 months before diagnosis, with some detectable as early as three years beforehand.
Pancreatic cancer is particularly difficult to diagnose because symptoms often emerge only after the disease has advanced, when treatment options may be more limited.
Unlike conventional approaches that primarily look for visible abnormalities, REDMOD examines microscopic-scale patterns reflected in CT scan data, potentially identifying cancer-related changes before a recognizable mass develops.
The researchers evaluated scans from patients who were eventually diagnosed with pancreatic cancer, alongside scans from individuals who remained cancer-free.
However, the technology is not foolproof.
In one independent test group, the system incorrectly flagged 81 of 430 people without pancreatic cancer, highlighting the possibility of unnecessary follow-up examinations and patient anxiety.
Promising, but not yet ready for hospitals
Despite its encouraging results, REDMOD remains an investigational tool rather than an approved cancer screening method.
Researchers must still determine how reliably it performs in real-world clinical settings, particularly among patients with elevated risk factors, including some individuals with newly diagnosed diabetes.
Further studies are also needed to establish whether earlier AI-assisted detection actually improves survival rates.
For now, the findings offer a promising direction for medical research rather than a replacement for doctors or existing diagnostic procedures.
The real breakthrough may not be teaching AI to see cancer better, but giving doctors more time to act before the disease becomes harder to treat.