Ivan Kairatov is a distinguished biopharma expert with a career dedicated to the intersection of medical research and technological innovation. With extensive experience in research and development, he has spent years navigating the complexities of drug discovery and diagnostic advancements, particularly focusing on how emerging technologies can solve the most stubborn challenges in oncology. His insights are grounded in a deep understanding of the regulatory and clinical hurdles required to bring life-saving tools from the laboratory to the patient’s bedside. In this discussion, we explore the groundbreaking potential of the REDMOD AI model, a next-generation tool developed at the Mayo Clinic designed to detect the subtle, nearly invisible precursors of pancreatic cancer long before they become lethal.
The conversation delves into the stark realities of pancreatic cancer survival rates and the clinical limitations that have historically made early detection nearly impossible. We examine how the REDMOD system utilizes radiomics to analyze tissue textures that escape the human eye, providing a lead time of over a year before a traditional diagnosis would occur. Furthermore, the dialogue covers the critical role of clinical red flags like new-onset diabetes, the challenge of maintaining diagnostic accuracy across different healthcare institutions, and the essential need for future studies to address demographic disparities to ensure equitable care for all patient populations.
Pancreatic cancer survival remains below 15% because 85% of cases are diagnosed after metastasis. How do these statistics shape current screening protocols, and what specific clinical challenges prevent traditional imaging from identifying these malignant lesions before they reach an unmanageable stage?
The grim reality of a five-year survival rate sitting below 15% casts a long shadow over the entire field of oncology, effectively turning a diagnosis into a race against a clock that has already been ticking for years. When you consider that 85% of patients only learn they have this disease after it has already migrated to other organs, it becomes clear that our current screening protocols are reactive rather than proactive. The primary clinical challenge is that the pancreas is tucked deep within the abdomen, and early-stage tumors often lack a discernible mass, making them “imaging-occult” or invisible to even the most seasoned radiologists. These lesions are frequently missed due to simple perceptual errors or technical limitations in standard CT resolution, where the texture of the malignant tissue is indistinguishable from healthy parenchyma. Until now, the medical community has been forced to wait for symptoms to manifest, but by then, the opportunity for curative intervention has almost always vanished.
Automated models can now identify subclinical changes 16 months before a clinical diagnosis, outperforming specialists by up to three-fold. Could you walk us through the specific tissue textures the technology analyzes and how it differentiates these subtle biological markers from benign abdominal findings?
The REDMOD model operates on a level of granular detail that the human brain simply isn’t wired to process, achieving an impressive area under the curve of 0.82 by identifying 73% of pre-diagnostic cancers. It focuses on filtered radiomic features—quantitative imaging markers that describe the “architecture” of the tissue, such as the heterogeneity of the cellular density and the specific way x-rays scatter through the organ. While a radiologist might see a uniform grey area, the AI detects micro-distortions and subtle biological shifts that indicate a transition toward malignancy at a median lead time of 16 months before any visible tumor appears. This is particularly striking when you look at scans taken more than two years before diagnosis, where the AI’s detection rate is nearly three times higher than that of specialists. It effectively separates these lethal signals from benign findings, like age-related changes or minor inflammations, by recognizing patterns of structural decay that are unique to the early stages of ductal adenocarcinoma.
New-onset diabetes and weight loss are often the first clinical red flags for pancreatic malignancy. In patients over 60 presenting with these symptoms, how does longitudinal monitoring of routine CT scans improve outcomes, and what steps should clinicians take when technology flags these invisible signs?
In the clinical world, glycemically-defined new-onset diabetes in a patient over 60 is like a silent alarm bell that should never be ignored, especially when paired with unexplained weight loss. The beauty of this AI-driven approach is its longitudinal stability; it can look back at scans from multiple institutions and imaging protocols to see if these “invisible” signs have been evolving over months or years. For a clinician, receiving a high-risk flag from a model like REDMOD means they no longer have to play a game of “wait and see” while a potential tumor grows. Instead, it prompts a risk-stratified evaluation, moving the patient into a high-intensity surveillance program or toward specialized diagnostic procedures that could lead to surgery while the cancer is still at stage 0. This shift from a single snapshot in time to a continuous, AI-augmented monitoring process is what will ultimately move the needle on those stagnant survival statistics.
Maintaining predictive accuracy across different imaging systems and protocols is a major hurdle in diagnostic technology. With specificity rates around 81%, how can medical teams balance these alerts with the risk of automation bias, and what criteria are necessary to prevent premature surgical interventions?
Achieving an 81% specificity is a significant milestone, but it also highlights the delicate tightrope we walk between early detection and the risk of over-treatment. The danger of “automation bias” is very real—the tendency for a tired or over-stressed physician to uncritically accept an AI’s alert—which is why the upcoming AI-PACED study is so vital for establishing override criteria. We must remember that an AI flag is a starting point for investigation, not a definitive mandate for the operating room. To prevent premature surgical interventions, medical teams need a tiered system where an algorithmic alert triggers secondary imaging, such as an endoscopic ultrasound, or specific molecular testing to confirm the AI’s suspicions. The goal is to create a “human-in-the-loop” workflow where the technology acts as a highly sensitive scout, but the final tactical decision remains a collaborative effort between the specialist and the data.
Racial and ethnic disparities often impact the risk levels associated with new-onset diabetes and cancer outcomes. How should future validation studies address these demographic variables, and what metrics are needed to ensure that early detection technology performs equitably across diverse patient populations?
We cannot ignore the fact that pancreatic cancer risk and diabetes outcomes are not distributed equally across all communities, and our technology must reflect that reality to be truly effective. The current validation of REDMOD is an incredible leap forward, but the next phase of research must intentionally include diverse datasets to ensure that the “invisible” markers being identified are consistent across different racial and ethnic groups. Metrics for success must go beyond just overall accuracy; we need to measure “algorithmic fairness” to ensure the 16-month lead time is equally accessible to a patient in an underserved urban clinic as it is in a major academic medical center. Future studies must prioritize the inclusion of populations that have historically been underrepresented in medical research to ensure that the AI doesn’t inadvertently perpetuate existing healthcare disparities.
What is your forecast for AI-driven pancreatic cancer screening?
I believe we are entering an era where the term “imaging-occult” will become a relic of the past as AI transforms standard abdominal CTs into powerful screening tools. Within the next decade, I forecast that AI models like REDMOD will be integrated directly into the digital infrastructure of every major hospital, acting as a silent sentinel that reviews every routine scan for the subtle textures of early-stage malignancy. We will likely see a significant shift where pancreatic cancer is no longer detected in the emergency room during a crisis, but rather in a primary care setting, months or even years in advance. As we refine these tools through prospective studies like AI-PACED and expand their reach to diverse populations, we will finally see that five-year survival rate climb out of the single digits, saving thousands of lives by catching the “invisible” before it becomes invincible.
