Mayo Clinic AI detects pancreatic cancer up to 3 years before diagnosis in landmark validation study
Artificial intelligence is changing the game in medicine, and the latest proof comes straight from the Mayo Clinic. A study published in the scientific journal Gut in April 2026 presents results that could completely transform how we deal with one of the deadliest cancers in the world.
The REDMOD model — short for Radiomics-based Early Detection Model — can identify signs of pancreatic cancer in routine CT scans up to three years in advance. That is before any tumor even becomes visible on imaging. It sounds like science fiction, but the data is real and has been rigorously validated.
Pancreatic cancer is one of the hardest cancers to treat precisely because, most of the time, it is only discovered after it has already spread throughout the body. More than 85% of diagnoses come at this late stage, when curative treatment options are already very limited. The five-year survival rate still sits below 15%, according to the National Cancer Institute — a scenario that, for a long time, seemed almost impossible to change. To make matters worse, projections indicate that pancreatic cancer is expected to become the second leading cause of cancer death in the U.S. by 2030.
But early detection may be about to gain a powerful ally. The study results show that REDMOD identified 73% of pre-diagnostic cancers with a median of approximately 16 months before clinical confirmation — nearly double the detection rate achieved by specialists analyzing the same scans without AI support. 🤯
And the earlier the scan was taken, the greater the model’s advantage. In CT scans performed more than two years before diagnosis, the artificial intelligence found nearly three times more cases than radiologists could detect on their own. We are talking about a massive leap in the ability to spot the disease when it can still be cured.
A robust study with data from multiple institutions
One of the most significant aspects of this work is that it was not conducted under artificial lab conditions. The researchers analyzed nearly 2,000 CT scans, including exams from patients who subsequently received a pancreatic cancer diagnosis — and all of those scans had originally been interpreted as normal. In other words, experienced radiologists looked at these images and saw nothing wrong, but REDMOD was able to pick up changes that were there, hidden in the data.
On top of that, the model was validated using data and workflows that mirror real clinical practice. This includes CT scans from multiple institutions, acquired with different imaging equipment and following varied protocols. This diversity in the dataset is crucial because it demonstrates that REDMOD does not only work under controlled conditions or with a specific type of scanner. It maintained consistent performance regardless of where the exams came from, which significantly increases confidence in its potential for large-scale application.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
Another important finding is that the model’s predictions remained stable over time. In patients who underwent multiple CT scans at different periods, REDMOD produced consistent results with months of interval between exams. This reinforces its usefulness for longitudinal monitoring, enabling clinicians to follow high-risk patients over the years and detect progressive changes in pancreatic tissue.
How REDMOD sees what the human eye cannot
The secret behind REDMOD lies in a field called radiomics, which is essentially the extraction and analysis of a massive volume of quantitative data from medical images. The model measures hundreds of quantitative imaging features that describe tissue texture and structure. While an experienced radiologist can identify visual patterns based on their training and years of practice, an artificial intelligence model trained on thousands of exams can capture extremely subtle variations in the texture, shape, and density of pancreatic tissue — changes that are virtually invisible to the human eye but already indicate that something is shifting in that organ well before any nodule or mass forms.
As Dr. Ajit Goenka, senior author of the study and a radiologist and nuclear medicine specialist at the Mayo Clinic, explained, the biggest barrier to saving lives from pancreatic cancer has been the inability to see the disease when it is still curable. According to him, this AI can now identify the signature of cancer from a normal-appearing pancreas, and it does so reliably over time and across different clinical settings.
In practice, the model analyzes CT scans that were already taken for other reasons — such as generic abdominal pain or a routine checkup — and can flag patients with an elevated risk of developing pancreatic cancer in the following years. This is especially relevant for high-risk patients, such as those with new-onset diabetes, a condition already known to be associated with an increased risk of pancreatic cancer.
The most interesting part is that REDMOD does not require any new exams, no additional infrastructure, and no time-consuming manual preparation. The model runs automatically on data that already exists, turning ordinary images into a highly sophisticated screening tool. This kind of passive screening approach has enormous scalability potential, especially in healthcare systems that already perform large volumes of imaging exams every day.
What this technology changes in practice
To understand the real impact of REDMOD, it helps to think in concrete numbers. Pancreatic cancer diagnosed at stage I has a five-year survival rate that can reach over 80%. When the diagnosis happens at stage IV — which is exactly the most common scenario today — that rate plummets to less than 3%. This staggering difference in prognosis is what makes early detection so critical for this type of cancer, and also what places REDMOD at the center of one of the most important discussions in oncology in recent years.
If the model can move the diagnosis forward by months or even years, it potentially shifts patients from a devastating prognosis bracket to one where treatment still has a real chance of a cure. It is no exaggeration to say that we are talking about the difference between life and death for many people.
Additionally, the possibility of integrating REDMOD into the normal workflow for analyzing CT scans means that high-risk patients could be identified and referred for specialized follow-up without any disruption to current processes. The model would function as an additional layer of intelligence on top of exams that already exist, generating alerts for the radiologists or attending physicians whenever it detects those risk patterns.
This does not replace healthcare professionals — quite the opposite, it amplifies their ability to make more informed decisions in a timely manner. When we are talking about a type of cancer with such a narrow treatment window, every month of lead time on the diagnosis can represent a massive gain in survival.
Another point that deserves attention is the impact on healthcare system costs. Treating pancreatic cancer at an advanced stage is extremely expensive and involves complex procedures, prolonged hospital stays, and palliative care that consumes significant resources. Early detection, on the other hand, opens the door for less invasive surgical interventions and more efficient treatment protocols. This is not just a matter of saving lives — although that is obviously the primary goal — but also of allocating healthcare resources more intelligently, which benefits the system as a whole.
AI-PACED: the path to clinical practice
Researchers at the Mayo Clinic are already advancing this work into the clinical trial phase through a study called AI-PACED — short for Artificial Intelligence for Pancreatic Cancer Early Detection. It is a prospective study evaluating how physicians can integrate AI-guided detection into the care of high-risk patients.
AI-PACED combines AI analysis of routine imaging exams with longitudinal patient follow-up to evaluate the model’s performance under real-world conditions. This includes measuring the early detection rate, the incidence of false positives, and the clinical outcomes of monitored patients. This is exactly the kind of rigorous, real-world validation that the medical community needs to see before adopting any new technology at scale.
This research is part of the Mayo Clinic’s Precure initiative, which aims to predict and prevent diseases by identifying the earliest biological changes in the body before symptoms begin. It is also aligned with the institution’s Clinical Impact strategy, which seeks to accelerate the translation of scientific discoveries into real patient care.
The study received funding from the National Institutes of Health, the Hoveida Family Foundation, the Mayo Clinic Comprehensive Cancer Center, and the Champions for Hope Pancreas Cancer Research Program at the Funk-Zitiello Foundation — a group of supporters that reflects the importance and credibility of the project.
The challenges that still need to be addressed
Despite the impressive results, the researchers are careful to emphasize that the model still needs to undergo additional validations across more diverse populations before it can be adopted at scale in clinical practice. The current study used data from multiple institutions, which is already a significant step forward, but it is essential to ensure that REDMOD’s performance remains consistent when applied across different ethnicities, age groups, health profiles, and CT equipment from various regions around the world.
Another important challenge lies in defining what happens after the model issues an alert. If REDMOD flags a patient as high risk, what is the ideal follow-up protocol? How frequently should that patient be reassessed? What additional tests should be ordered? These questions are still being discussed by the medical community, and the answers will depend on both clinical evidence and practical considerations about feasibility within different healthcare systems.
The issue of false positives also deserves attention. Any screening tool needs to balance sensitivity and specificity. Detecting many cases early is great, but if the model generates an excessive number of false alarms, it could overwhelm the healthcare system with unnecessary tests and procedures, in addition to causing significant anxiety for patients. That is why the AI-PACED study is specifically measuring this variable — to understand how the model behaves in the real world and adjust alert thresholds as needed.
What this means for the future of medicine
What is clear is that artificial intelligence is opening doors that seemed locked just a few years ago. REDMOD represents a genuine breakthrough, not only because of its technical performance, but because of what it symbolizes: the real possibility of transforming routine scans into an early warning network against one of the most lethal cancers in existence. 🧬
REDMOD’s approach also raises a broader reflection about the role of AI in healthcare. We are not talking about replacing doctors with algorithms. We are talking about giving healthcare professionals a diagnostic superpower — the ability to see what is hidden in the data, to catch patterns that no human eye, no matter how well-trained, could consistently identify across thousands of exams. It is a partnership between human expertise and computational capability, and when that partnership works well, the patient wins.
Pancreatic cancer has always been considered one of the great villains of oncology, precisely because of its silent and aggressive nature. But tools like REDMOD show that the era of passive and late detection may be coming to an end. If the upcoming clinical studies confirm the current results, we are looking at a technology that could save hundreds of thousands of lives per year around the world — and that, without a doubt, is one of the most important stories that medicine and artificial intelligence have to tell right now.
