AI can now diagnose the "silent killer" pancreatic cancer in advance.

Mayo Clinic research team developed an AI model called Redmod, claiming it can identify subtle changes undetectable by the human eye from routine CT scans an average of 475 days before pancreatic cancer diagnosis. The study shows an overall AI detection rate of 73%, compared to 39% by radiologists; for scans more than two years prior, the gap widens to 68% versus 23%.
(Background: AI “cancer screening” accuracy reaches 98%! Cambridge study shows DNA alone can accelerate early diagnosis and treatment)
(Additional context: Anthropic invests $400 million to acquire AI biotech startup Coefficient Bio, directly competing with OpenAI)

Table of Contents

Toggle

• The 475-day window
• AI sees it, doctors do not
• Doubling survival rates

The most heartbreaking issue with pancreatic cancer is often discovering it too late. Over 85% of cases worldwide are diagnosed at an advanced stage, with tumors no longer operable; the five-year survival rate hovers around 10%, one of the bleakest figures among common cancers.

However, a recent study published at the end of April in the medical journal Gut attempts to address the root problem: not waiting for symptoms to appear, not waiting for tumors to become visible, but enabling AI to find early signals invisible to the human eye in routine CT scans.

The 475-day window

The Mayo Clinic research team developed an AI model called Redmod, aiming to identify subtle changes in routine abdominal CT images that radiologists cannot detect. The training dataset included over 1,400 subjects, notably a key group of 219 patients who were initially deemed “normal” on scans but later diagnosed with pancreatic cancer.

Redmod AI reanalyzed these seemingly normal images from years prior and concluded it could identify abnormal features an average of 475 days before diagnosis.

475 days is about one year and four months. During this period, tumors are usually still localized and resectable. Clinically, this means: a patient with no symptoms yet could be identified before surgical intervention is no longer possible.

The core dilemma of pancreatic cancer lies in its “silence”: early stages do not cause symptoms nor appear on imaging. By the time patients seek medical attention due to discomfort, the tumor has often already breached regional boundaries. Redmod AI aims to intervene during this silent phase.

AI sees it, doctors do not

The study compared AI and radiologists head-to-head on the same set of images. The results: Redmod’s overall correct detection rate was 73%, while radiologists achieved 39% — nearly a twofold difference.

For scans taken more than two years before diagnosis, the gap is even more pronounced: Redmod identified 68% of cases, while radiologists only 23%. In other words, when a tumor is still more than two years away from being “visible,” AI’s detection capability is three times that of human doctors.

The model’s specificity is also noteworthy: in control images where no cancer developed, Redmod correctly classified over 80%. It is not only trying to catch all cancers but also avoids false positives.

The study also confirmed that the model performs stably across different hospitals and CT device brands, meeting basic deployment criteria.

Doubling survival rates depends on this

The researchers cite a modeling estimate: if the proportion of localized pancreatic cancer cases—meaning tumors that have not yet spread—can be increased from the current 10% to 50%, the five-year survival rate could more than double.

The logic is straightforward: early diagnosis is not just about knowing bad news sooner, but about significantly increasing the chances that patients can undergo curative surgery.

However, the researchers explicitly state that Redmod still requires prospective clinical trials: continuous tracking of patient outcomes in real screening scenarios to confirm it truly improves survival rates before routine use. Retrospective data showing excellent performance is necessary but not sufficient.

The team’s near-term application plan is to target high-risk groups: older patients, those with unexplained weight loss, and new-onset diabetes. This is a precise and feasible entry point, not large-scale population screening, but leveraging existing clinical information to help AI identify who needs closer monitoring.

The researchers wrote in the paper that the 475-day window “has profound significance,” because within this period, cure is not an exception but a possible norm. Whether AI can see this early enough to make subsequent treatment meaningful is the next question.