Artificial intelligence (AI)-enabled medical devices are advancing at an unprecedented pace—from software that helps interpret medical images to detect cancerous growths or identify biomarkers for specific diseases to algorithms that analyze ECGs for potentially fatal heart arrhythmias—unlocking new opportunities for care while also raising regulatory questions about how to evaluate safety and performance over time.
The University of Maryland’s College of Computer, Mathematical, and Natural Sciences (CMNS) is launching a new research collaboration with the U.S. Food and Drug Administration’s Center for Devices and Radiological Health to develop new methods for assessing the reliability of AI- and machine learning-enabled medical devices before and after deployment.
“This new partnership reflects our commitment to solving grand challenges and maximizing the real-world impact of our research,” said Jennifer King Rice, senior vice president and provost at UMD. “By working together with the FDA at the forefront of regulatory science, our faculty will help new AI technologies reach patients with the safety, performance and trust they deserve.”
Through this collaboration, researchers at the University of Maryland Institute for Health Computing (UM-IHC) in North Bethesda, Maryland, will benefit by partnering with the FDA on joint projects aimed at translating research into tools and frameworks that can support regulatory decision-making.
“Our goals are to develop techniques to assess the safety and effectiveness of new AI-enabled medical devices and to monitor performance after deployment,” said the collaboration’s UMD principal investigator Amitabh Varshney, a professor of computer science and dean of CMNS.
“AI-enabled medical devices are transforming the availability, accuracy and timeliness of medical information, and this collaboration aligns with our focus at the IHC on advancing evidence, methods and standards that keep pace with innovation,” added Adam Porter, co-executive director of UM-IHC and a professor of computer science at UMD.
One focus area of the collaboration is developing ways to measure and communicate how AI systems reach their results. The team will study quantitative “explainability” measures designed to help determine whether model outputs are clinically meaningful and reliable. Key metrics include fidelity (accuracy to the model), plausibility (alignment with medical knowledge), consistency (stability) and usefulness (impact on clinical tasks).
Another focus is performance change over time, such as model or data drift, which can occur when real-world data differs from the data used to train a model. Over time, this can degrade the device’s accuracy and reliability, but there are currently no standardized tools or datasets for monitoring AI systems after deployment.
“We will work together to identify mechanisms that can explain the causes of model drift,” Varshney added. “Understanding why model performance changes may enable informed adjustments to maintain the accuracy of these systems in real-world use.”
The team will also develop tools and metrics to support the evaluation of AI-enabled extended reality (XR) devices for clinical and health care settings, where some systems can be limited by cybersickness during prolonged use. The researchers plan to refine and further validate an existing FDA-cleared predictive algorithm for cybersickness risk.
“The resulting methodology and enhanced algorithm will provide objective, reproducible evidence needed to support accelerated regulatory approval pathways for XR-based medical interventions,” Varshney said.
In the coming days, a call for proposals will be issued to UM-IHC faculty members, and two projects are expected to be funded this year.
