AI Doctor MIRA Completes Full Workflow: Demonstrating Action Capability in Clinical Settings
Nature: AI医生MIRA首次“全流程上岗” - 虎嗅网
Summary
Based on a Nature paper, the autonomous medical AI agent 'MIRA' was developed. This system demonstrated its ability to autonomously execute an entire workflow—from patient history taking to diagnosis, testing, and treatment planning—within an EHR sandbox environment.
Details
The study addresses the limitation of traditional LLMs being mere 'answerers,' demonstrating that AI can become an 'acting agent' in real clinical settings. MIRA is innovative because it utilizes an isolated EHR sandbox environment compatible with FHIR protocols and standardized coding systems (ICD/LOINC/SNOMED-CT), allowing it to output structured data commands for diagnoses and prescriptions. Using MIMIC-IV database emergency cases, the system simulated end-to-end workflows—from history taking to admission. Results showed that MIRA maintained high diagnostic accuracy (87.8%) comparable to specialists and did not exhibit excessive testing or high-risk medication errors. However, the authors conclude that since this was a simulation in an isolated environment, further validation and the establishment of human oversight mechanisms are essential for real-world application.
Original content copyright by respective publishers