AMIE video consultations matched physicians in a controlled study
Google’s research medical AI system AMIE (Video) was assessed in synchronous video consultations with 15 trained patient actors. The actors presented cases involving cardiopulmonary, abdominal, HEENT, neurological or psychiatric, and musculoskeletal conditions. An independent panel of 20 experienced primary care physicians reviewed the sessions, alongside comparisons with 10 board-certified primary care physicians and a text-only AMIE version. Google reported that evaluators rated the video system comparably to the physician group for history-taking, diagnostic accuracy, management recommendations, and communication. Patient actors preferred video to text and gave AMIE favourable ratings for empathy, rapport, and confidence in care.
AMIE uses three agents: a talker manages the spoken exchange, a planner updates possible diagnoses and management plans, and a perception agent continuously examines audio and video for relevant signs. Separating these functions is intended to reduce delays caused by detailed reasoning and visual processing.
The findings came from prepared scenarios with professional actors, not real patients. Google noted perception and reasoning errors, occasional technical interruptions, and limits in reproducing real-world variability. The company said studies with real patients are still needed before conclusions about clinical use can be drawn. Related text-based AMIE work includes studies with Beth Israel Deaconess Medical Center and Included Health.
Written by Sakhanda Wire AI from reporting by AI News. Facts, quotes and full context are in the original article below.