Dresden (LabNews Media LLC) – Researchers from the Else Kröner Fresenius Center for Digital Health at the Technical University of Dresden and the Dresden University Hospital have developed an autonomous AI agent that can independently perform clinical tasks in electronic patient records. The system, named MIRA, achieved higher diagnostic accuracy than the physician comparison groups in retrospective simulations with real patient cases, while simultaneously making guideline-compliant and safe treatment decisions.
MIRA (Medical Intelligence for Reasoning and Action) was programmed to identify missing information, request examinations, interpret findings, and prepare treatment decisions within a protected test environment. In the tests, the agent used a controlled clinical toolkit with eleven instruments and over 85,000 possible action options, including laboratory, microbiological, and imaging examinations, as well as medication prescriptions and admission decisions.
For the study, more than 500 real patient cases were recreated in a simulated emergency room. MIRA interacted with virtual patients whose responses were based on the documented medical histories from the original records. This allowed the system to take a medical history, ask targeted questions, and incorporate the information obtained into clinical decision-making.
The results show that AI agents like MIRA could take over routine tasks as clinical co-pilots in the future and relieve medical professionals. However, the responsibility for the final decisions will continue to lie with the medical staff.
The study was published in the journal Nature. The research team led by Professor Jakob N. Kather sees MIRA as a preview of how AI systems could be integrated into clinical practice in the future.
