Published
ARPA-H launches program to help AI-enabled medical tools maintain peak performance
Program aims to develop techniques to automatically identify and correct performance degradation of AI-enabled clinical decision support tools
The Advanced Research Projects Agency for Health (ARPA-H), an agency within the U.S. Department of Health and Human Services (HHS), today announced a new funding opportunity through the Performance and Reliability Evaluation for Continuous Modifications and Useability of Artificial Intelligence (PRECISE-AI) program. PRECISE-AI intends to develop capabilities that can detect when an AI-enabled tool used in real-world clinical care settings is potentially out of alignment with its underlying training data and, importantly, auto-correct these AI-enabled tools to maintain their peak performance, making them more helpful to clinicians and beneficial to patients.
Today, more than 950 medical devices with integrated AI functionalities have been authorized by the U.S. Food and Drug Administration (FDA)—a tenfold increase from 2018 to 2023. These new AI-enabled tools transform doctors’ ability to provide care, from virtual assistants to enhanced medical imaging and diagnostic tools. However, research suggests that Machine Learning models (ML) used in medical AI-enabled tools may degrade over time due to changes in input data—such as changes in clinical operations, data acquisition, patient population, or even IT infrastructure. Current manual techniques make it challenging to monitor and maintain the performance of these AI-enabled tools in real-world settings.
“The promise of AI-enabled tools for health care is only as strong as the relevant real-world data informing them,” said PRECISE-AI Program Manager Berkman Sahiner. “The inability to automatically monitor and maintain an AI-enabled medical device’s performance based on real-world operations creates risks for clinicians and patients. PRECISE-AI aims to create a suite of techniques that can maintain peak performance of AI-enabled tools by recommending the optimal approach to detect and mitigate the underlying AI model performance degradation so that clinicians can provide better patient care.”
PRECISE-AI will develop techniques that analyze AI-enabled tools and identify root causes for performance deterioration. The root cause analysis will inform self-corrective actions to improve the AI-enabled tool’s performance. The program will also create mechanisms for notifying clinicians, AI tool developers, hospital administrators, and regulators when performance degradation occurs.
“PRECISE-AI is addressing a growing gap in ensuring AI tools used in clinical decision-making are accurate, safe, and robust in real-world settings,” said ARPA-H Director Renee Wegrzyn, Ph.D. “In doing so, ARPA-H is creating a foundation of trust between clinicians and these AI tools, which will further expand AI’s potential in improving health outcomes for all Americans.”
Through a forthcoming program solicitation, PRECISE-AI will bring together ML experts, health information specialists, and clinicians to develop novel techniques across five technical areas (TAs). Performer teams will start by developing tools to establish the most accurate estimate of a patient’s diagnosis given the available evidence or the “ground truth.” Teams will then develop autonomous methods to monitor an AI tool’s performance against these “ground truths,” determine the root causes of any degradation, and make necessary corrections. Additionally, teams will develop methods for the underlying AI model to communicate uncertainty to clinicians and other users, as well as develop the data infrastructure to allow findings to be shared with other clinicians, developers, and hospital administrators using the AI-enabled tool. The progress of all the performers will be confirmed by an independent verification and validation team.
Multiple awards under this solicitation are anticipated. Awards will depend on the quality of the proposals received and the availability of funds. Learn more about PRECISE-AI on its program page, including information about Proposers’ Day registration and guidance on submitting a teaming profile.