Microsoft develops AI tool that can diagnose ‘sequentially’ like a doctor

Microsoft researchers have developed an AI tool capable of diagnosing patients step-by-step, mirroring the approach of doctors, according to a study published in Science on Thursday. The AI model outperformed general physicians in solving complex medical cases, demonstrating a more than twice accurate rate, as reported by Dr. Dominic King, corporate vice president of the health unit at Microsoft AI.

The model, equipped with the "diagnostic orchestrator," determines the sequence of questioning patients, ordering relevant exams and tests, all while dynamically updating its working hypotheses and plans. Unlike traditional medical AI models which follow an all-at-once approach, the orchestrator simulates a panel of clinical experts, optimizing the diagnostic process.

The tool not only proved more accurate when tasked with diagnosing over 300 rare and perplexing medical dilemmas but also performed better than human physicians in scenarios sourced from a major database of emergency room admissions. Furthermore, the orchestrator's diagnostic work-ups were found to be more cost-effective than the conventional methods employed by human doctors, as highlighted by Harsha Nori, head of AI at Microsoft AI Health.

The researchers, in collaboration with Beth Israel Deaconess Medical Center in Boston, plan to further assess the tool's effectiveness in triaging patients in emergency department waiting rooms.