Mayo Clinic has roughly 150 AI models across its health system, covering everything from AI scribes and early cancer detection to ECG analysis, pathology and emergency-department triage. Some are already used in clinical practice, while others are still being validated or developed. They all require computing power, data and robust governance.
Up to 30 minutes less time spent analyzing patient records
Mayo Clinic knows everyone in the US. The health organization, which treats 1.3 million people from more than 130 countries annually, has been developing and implementing artificial intelligence in administration, diagnosis and treatment, and clinical research for several years. Today, its portfolio of AI models has reached roughly 150, according to a recent CNN interview with Matthew Callstrom, medical director of the generative AI program at Mayo Clinic.
Mayo does not officially publish a complete list of all its AI models, and the portfolio includes applications at different stages of development, validation and deployment. However, the most important ones have been revealed in interviews, scientific publications and cooperation announcements.
The recently revealed Record Time is an AI-powered solution developed in collaboration with Scale AI. It was created to address fragmented medical records. Patients often provide medical records from multiple other health systems. In complex cases, these records can amount to several hundred pages. Record Time organizes the information chronologically and helps physicians identify the most relevant facts.
According to Mayo, the solution can reduce appointment preparation time by 5–30 minutes per patient. AI takes over part of the work of searching for and organizing information, while diagnosis and treatment decisions remain the physician's responsibility. The solution is currently being tested.
AI searches for hidden diseases in ECGs and signs of pancreatic cancer
AI-ECG is a family of AI models developed by Mayo Clinic to identify patterns in standard electrocardiograms that may indicate otherwise difficult-to-detect heart conditions.
Mayo has been developing AI-enhanced ECG models for several years. One application is designed to identify patients at risk of cardiac amyloidosis, a disease that can be difficult to diagnose, particularly in its early stages. The model was developed using ECG data collected from 2000 to 2019.
Other Mayo AI-ECG models have been developed to detect conditions including low ejection fraction, atrial fibrillation, aortic stenosis and hypertrophic cardiomyopathy. A Mayo AI-ECG algorithm for detecting low ejection fraction received FDA clearance for clinical use in 2023. Other applications, including those designed to identify cardiac amyloidosis, are still being studied.
Another diagnostic model, REDMOD, uses computed tomography (CT) image analysis for early detection of pancreatic cancer. Rather than simply looking for an already visible tumor, the model analyzes subtle changes in the structure of the pancreas that may precede conventional diagnosis.
In a validation study, REDMOD identified approximately 73% of prediagnostic pancreatic cancers, at a median of about 16 months before diagnosis. Its detection rate was nearly twice that of specialists reviewing the same scans without AI assistance. The model is still undergoing further clinical validation.
AI in pathology and palliative care
In collaboration with Aignostics and Charité – Universitätsmedizin Berlin, Mayo has developed Atlas, a pathology foundation model trained on more than 1.2 million histopathology whole-slide images (WSIs). Rather than being a single diagnostic application, Atlas is a foundation model that can support the development of different AI applications for pathology.
At the same time, Mayo is developing a digital work environment for pathologists. Fusion AP, developed with Techcyte, integrates a histopathology image viewer with clinical data and AI tools.
The platform gives pathologists access to more than 22.6 million whole-slide images, including current and historical specimens. This lets images be viewed alongside related clinical information and provides infrastructure to deploy and evaluate additional AI applications in digital pathology.
AI is also supporting palliative care. In collaboration with Bayesian Health, Mayo developed a system that analyzes data on hospitalized patients and helps identify those who may benefit from an earlier palliative care consultation. In a clinical trial, use of the system was associated with a 44% increase in timely referrals to palliative care. The study also found reductions in 60- and 90-day readmissions. The system is integrated into the electronic health record, surfacing information about a patient's potential need for palliative care within the clinical workflow.
Precision medicine, radiology and disease prediction
Together with Cerebras, Mayo Clinic is developing the Mayo Clinic Genomic Foundation Model, which uses AI to analyze genetic variants and genomic data. The initial focus is rheumatoid arthritis, aiming to improve prediction of treatment response and support more personalized treatment selection. The model remains in development and is not currently being used as a routine clinical tool.
In collaboration with Microsoft Research, Mayo is also developing multimodal AI models for medical imaging. One application being explored involves chest X-rays. Potential uses include generating preliminary reports, checking catheter placement, and comparing a patient's examinations over time.
The number of predictive AI applications is also growing. Mayo and Predicate AI, for example, have worked on solutions designed to identify the risk of sepsis at an earlier stage using information from patient monitoring devices and electronic health records.
AI is being explored in hospital operations, including triage, patient-flow management, bed-demand forecasting and surgical scheduling. Researchers at Mayo Clinic Arizona have developed a model that uses emergency department triage data to predict whether a patient is suitable for a vertical processing pathway (VPP). Under this approach, selected patients can remain seated or standing rather than being placed in a traditional hospital bed. The model was trained using data from 49,350 emergency department visits.
Another 2026 study by Mayo researchers examined an AI model designed to predict, at the time of emergency department triage, whether a patient will require hospitalization. Mayo researchers are also developing AI tools that analyze patient-submitted photographs of postoperative wounds. The aim is to identify possible surgical-site infections and help clinical staff determine which cases require further attention.
All these models require energy, data, computing power and governance
A portfolio of roughly 150 AI models requires substantial computing infrastructure. Mayo has long relied on cloud technology. In 2019, it entered into a 10-year strategic partnership with Google centered on Google Cloud, which serves as an important part of Mayo's digital infrastructure.
In June 2026, Mayo and Microsoft announced a collaboration to develop a frontier AI model specifically for healthcare. The model is intended to support a broad range of clinical reasoning and healthcare use cases rather than function as a “ChatGPT for doctors.” Microsoft plans to make the model available through Azure Foundry APIs.
At the same time, Mayo has invested in its own high-performance computing infrastructure. In 2025, the organization deployed an NVIDIA DGX B200 SuperPOD to support the development of foundation models and other AI applications, including work in digital pathology, drug discovery and precision medicine.
Mayo's approach to AI is not simply to develop a model and immediately put it into clinical use: The organization has established an AI governance and lifecycle-management process under which clinical AI applications undergo review before Mayo staff can use them.
The assessment can include model performance, patient safety, workflow integration, privacy, security, and lifecycle management. Depending on the application, validation may involve retrospective studies, prospective clinical research or other forms of evaluation. Not every AI model necessarily goes through a clinical trial.
Once deployed, applications are subject to ongoing performance and lifecycle monitoring. This is particularly important because an AI model that performs well during development can behave differently when exposed to new patient populations, workflows or data.
As a result, Mayo is becoming a major research and development center for medical AI. The organization has also developed Mayo Clinic Platform_Connect, a federated health data network that enables organizations to develop, test and validate algorithms using secure, de-identified clinical data while participating institutions retain control over their own data. The goal is to create an infrastructure that combines clinical data, computing power, research expertise, and clinical validation.
References
Photo: Shutterstock
