ARPA-H Launches Program to Maintain AI-Driven Medical Tools

The Advanced Research Projects Agency for Health (ARPA-H) has initiated a program aimed at preserving the performance of medical tools utilizing artificial intelligence.

A part of the U.S. Department of Health and Human Services, the agency announced on Thursday that it will provide funding to identify and auto-correct AI-enabled tools that may deviate from their original training data.

This program is based on research indicating that machine learning models‘ accuracy might deteriorate over time due to changes in variables such as clinical processes, data collection, and patient demographics.

Details of the Program

The Food and Drug Administration has approved 950 medical devices with AI/machine learning capabilities integrated. Despite evidence showing accuracy degradation over time, no routine testing is conducted on these models during clinical usage to ensure their continued expected performance, and there are no mandates for updating underperforming AI.

According to ARPA-H, „Clinical intuition by the physician employing the technology“ is the primary means to detect AI model degradation. However, this method can be inconsistent and unreliable, potentially leading to misdiagnosis before any issues are noticed. The agency highlighted a lack of technical solutions to address this challenge.

To tackle this issue, the ARPA-H has introduced the Performance and Reliability Evaluation for Continuous Modifications and Useability of Artificial Intelligence (PRECISE-AI) program.

The program aims to develop tools that can monitor AI models‘ performance, identify degradation, and autonomously correct performance issues without human intervention, alleviating the burden on operators. The agency also emphasized the necessity for clear communication of degradation sources.

Implementation and Goals

The four-year PRECISE-AI program comprises five technical areas, covering activities like continuous performance monitoring and uncertainty communication. Participants are expected to spend the initial two years prototyping their approaches and the subsequent two years testing them in real-world scenarios and integrating them commercially.

This initiative serves as an early examination of ARPA-H, established by the Biden administration in 2022. Modeled after the Defense Advanced Research Projects Agency, ARPA-H’s mission is to advance healthcare areas that traditional research or commercial activities cannot accomplish.