Mayo Clinic Partners with Microsoft and Cerebras on Radiology and Genomics Foundation Models

Mayo Clinic has announced collaborations with Microsoft Research and Cerebras Systems to explore the potential of generative artificial intelligence (AI) in patient care, diagnostics, and treatment planning. These projects, introduced during the 43rd Annual J.P. Morgan Healthcare Conference, focus on developing foundation models tailored for medical use by integrating diverse datasets, including radiology images and genomic sequencing data.

Cerebras Systems, specializing in AI hardware, is best known for its Wafer Scale Engine (WSE), a dinner plate sized chip, designed to accelerate deep learning applications. Their third-generation WSE-3, introduced in 2024, features 4 trillion transistors and 900,000 AI-optimized cores, capable of training models with up to 24 trillion parameters.

Mayo Clinic and Microsoft Research are working together to enhance radiology workflows by combining X-ray data with generative AI technology. The collaboration aims to create tools capable of automating report generation, assessing tube and line placement, and detecting changes in sequential chest X-rays.

Matthew Callstrom, M.D., Ph.D., chair of Mayo Clinic Radiology in the Midwest and medical director for Generative AI and Strategy:

"Foundation models combining text and images address critical challenges in radiology. This collaboration seeks to develop tools that enhance imaging workflows and clinical decision-making. Our focus on chest X-ray reports allows us to explore practical AI applications in radiology."

In another collaboration, Mayo Clinic and Cerebras Systems have developed a genomic foundation model using the Cerebras Wafer Scale Cluster, leveraging publicly available human genome data alongside Mayo’s patient exome data. This model is designed to support diagnostics and treatment planning, with an initial focus on rheumatoid arthritis (RA). The genomic model, trained with one trillion tokens and featuring one billion parameters, has demonstrated high accuracy in clinical evaluations, including predictions of cancer predisposition (96%), cardiovascular risk (83%), and RA drug response (87%).

The genomic model is designed to analyze genomic data in real time, compare it with patient cohorts, and predict disease patterns. Initial results suggest it may reduce the time required to identify effective treatments for RA, lowering risks associated with delayed interventions.

Cerebras Systems is known for its AI hardware innovations, including the Wafer Scale Engine, the world’s largest AI processor, which accelerates computational tasks for AI and machine learning models. Their technology powers the genomic foundation model in this collaboration. Cerebras Systems previously collaborated with M42 to train and fine-tune the Med42 Clinical Large Language Model to advance clinical decision-making and personalized treatment plans.

Topics: AI & Digital