A one-year pilot grant from U-M’s new biomedical and bio-innovation institute, known as Unit X, and a Google Research Award will provide funding and advanced computing resources for Dr. Wang’s work on AI-empowered precision medicine.
Two new awards will help University of Michigan Biomedical Engineering (U-M BME) Assistant Professor Joyce Yan-Ran Wang translate an emerging artificial intelligence framework into tools that could support complex clinical decisions.
A one-year pilot grant from U-M’s new biomedical and bio-innovation institute, known as Unit X, and a Google Research Award will provide funding and advanced computing resources for Dr. Wang’s work on AI-empowered precision medicine.
“The smart patient retrieval project is a direct response to our paper published in March in Nature Reviews Cancer,” Wang said. “We advocated for this area because we think it is the future of precision medicine.”
Tumor boards bring together oncologists, radiologists, pathologists and other specialists to interpret evidence and recommend treatment. As precision medicine advances, however, clinicians must consider increasingly complex combinations of medical images, pathology findings, molecular profiles, clinical records and treatment histories.
In the Nature Reviews Cancer perspective, “AI-driven smart patient retrieval for precision oncology,” Wang and her co-authors proposed systems that could search across these different types of data to identify previously treated patients who are clinically similar to a new patient. Clinicians could then review those patients’ treatments, responses and long-term outcomes as additional evidence during decision-making.
Unlike keyword searches or rigid database filters, smart patient retrieval would use semantic, multimodal AI models to recognize clinically meaningful similarities distributed across several data sources. The approach could expand upon a familiar form of clinical reasoning—“I had a similar patient who…”—by drawing on collective experience beyond any individual clinician or institution.
“Modern clinical decision-making increasingly requires the integration of highly heterogeneous data, including imaging, pathology, molecular profiles and longitudinal clinical records, which can be difficult to synthesize simultaneously,” Wang said. “AI, particularly multimodal representation learning, offers a way to integrate these disparate data types and identify clinically meaningful patterns at scale. Our goal is to develop an AI-enabled multimodal patient retrieval system that can identify prior patients with biologically and clinically meaningful similarities to a new patient and surface their treatment responses, outcomes and longitudinal disease trajectories. By grounding clinical decision support in retrieved, verifiable evidence from real-world patient records, the system can substantially reduce the risk of AI hallucination and improve the transparency, traceability and clinical reliability of its recommendations. Rather than generating conclusions in isolation, the AI is designed to anchor its reasoning in evidence from relevant prior cases, providing clinicians with an interpretable and evidence-based framework for complex decision-making.”
This capability could be particularly useful for rare cancers, unusual presentations and other complex diseases for which treatment guidelines may be limited or may not reflect an individual patient’s complete clinical profile.
The goal is not to replace clinicians or prescribe treatment automatically. Instead, the system would help clinical teams find relevant evidence and devote more of their time to deliberation.
Unit X is a new U-M biomedical and bio-innovation institute designed to bring together expertise from Michigan’s health, engineering and other schools and colleges. Coupled with a new supercomputing platform, the institute aims to accelerate biomedical discovery, expand translational research and attract entrepreneurial talent.
U-M President Domenico Grasso announced the institute during a recent State of the University address, describing an initial investment of $250 million over five years. He compared the institute’s ambitions to those of Lockheed Martin’s Skunk Works, which U-M graduate Kelly Johnson established during World War II to rapidly develop innovative aircraft.
“Our new institute will carry that same revolutionary spirit forward, advancing health sciences, bio AI, and innovation for the public good,” Dr. Grasso said.
The institute’s launch is being co-led by Arul Chinnaiyan, the S.P. Hicks Endowed Professor of Pathology, professor of urology and director of the Michigan Center for Translational Pathology, and Joerg Lahann, the Wolfgang Pauli Collegiate Professor of Chemical Engineering and director of the Biointerfaces Institute. Dr. Lahann also holds appointments in materials science and engineering, biomedical engineering, and macromolecular science and engineering.
Unit X is intended to provide resources for the broader biomedical community, establish a new model for creating and translating ventures, and recruit researchers who can expand U-M’s entrepreneurial faculty. Its emerging research platforms will support fields including bioartificial intelligence, generative biology and automated biological laboratories.
“Unit X represents a bold new chapter for Michigan—a place where cutting-edge biology, computation and engineering will converge to accelerate discoveries into real-world impact,” Dr. Chinnaiyan said.
Its emphasis on integrating AI with precision medicine and translational research aligns closely with Dr. Wang’s efforts to develop clinically grounded multimodal systems.
Dr. Wang’s one-year Unit X pilot grant will support the development of the smart patient retrieval system with approximately $200,000 in funding and computing resources. Support includes access to cloud-based graphics processing units and personnel funding to work on the project.
The award is part of Unit X’s support for projects at the intersection of AI, health care and biology. In addition to advancing individual studies, the pilot program will help assess the computing infrastructure that AI-enabled health research requires, including the types and quantities of graphics processing units researchers need and the respective roles of cloud and local resources.
A central research challenge will be determining what makes two patients meaningfully similar. Patients might share keywords or individual clinical features without being comparable in a way that is useful for care. Conversely, an important match might emerge only when the system considers a combination of imaging, clinical, pathological and molecular information.
“Defining similarity is the hard part,” Dr. Wang said. “Similarity is not just two patients matching on a chart or sharing some keywords. We need clinicians to work with us to define what a clinically meaningful match means.”
Dr. Wang’s lab is seeking collaborators in oncology, neuro-oncology, cardiovascular medicine, intensive care and other specialties. Their expertise will help the researchers develop similarity measures that reflect real clinical reasoning and workflow needs.
“We are actively looking for clinicians who are interested in this project or have relevant ideas,” Wang said. “We want to build an AI system that truly benefits clinical practice and augments clinicians’ capabilities.”
Her lab is also recruiting postdoctoral researchers with a computer science background, particularly experience with large language models and vision-language models.
The Google Research Award provides $100,000 in industry-sponsored research support to advance Dr. Wang’s broader portfolio of AI and health care research. The award strengthens the team’s capacity to pursue translational, data-intensive work at the intersection of artificial intelligence and clinical care.
Building on the Unit X pilot, Dr. Wang’s team plans to extend the framework of smart patient retrieval beyond identifying similar patients to predicting prognosis and adverse events from multimodal clinical data.
“Retrieving similar patients is an intermediate step,” Dr. Wang said. “The next question is whether an AI system can use multimodal data to predict a patient’s prognosis.”
For example, the researchers could investigate whether AI models can estimate a patient’s five-, 10- or 15-year risk of heart attack or ischemic stroke in the setting of cardiovascular disease and stenosis. These predictions could integrate information distributed across imaging studies, structured electronic health record data, clinical notes and longitudinal outcomes.
The Google Research Award complements institutional support from the Unit X pilot and expands the resources available for this research program. In particular, the industry-sponsored award provides additional high-performance computing capacity for training and evaluating large multimodal models, together with unrestricted funding to support research personnel and accelerate the development of new AI methods for clinical applications.
Both projects carry forward the roadmap outlined in Dr. Wang’s Nature Reviews Cancer perspective. That article emphasized that algorithmic similarity does not automatically equal clinical relevance and that real-world outcomes must be interpreted carefully. Safe deployment will require clinician co-design, rigorous validation, uncertainty estimates, bias assessment, audit trails and human review.
The authors also proposed that large language models could assist clinicians by creating structured case summaries and synthesizing complex patient information, provided their outputs are grounded in retrieved clinical evidence rather than unsupported generation.
For Dr. Wang, collaboration with clinicians remains essential at every stage—from defining meaningful patient similarity to assessing whether an AI system’s results are genuinely useful in practice.
“This is not only for oncology; it can extend to other complex diseases,” Wang said. “We want to work together with clinicians and researchers to move this project forward and develop AI that supports more focused, effective and individualized care.”