Microsoft AI tool outperforms doctors in complex diagnostic tests, study finds

Microsoft researchers discovered an AI tool that diagnoses patients step-by-step, as doctors do, outperforms general physicians in complex medical cases, according to a study published in Science. The model uses an orchestrator to coordinate questions, exams, and tests, imitating a simulated panel of clinical experts. Older medical AI models typically diagnose all symptoms, lab results, and imaging studies at once, while this new approach takes a more methodical process.

The orchestrator first determines what to ask about the patient's history and physical exam, then decides which tests to order, adjusts hypotheses, and updates plans accordingly. When confronted with over 300 unusual medical dilemmas, the AI model diagnosed patients more accurately than human physicians using textbooks and the internet, performing at twice the accuracy rate.

Additionally, the tool proved more cost-effective in designing work-ups compared to human doctors. The researchers, in collaboration with Beth Israel Deaconess Medical Center in Boston, plan to test the tool's capability to triage patients in emergency department waiting rooms.