Microsoft Develops AI Diagnostic System That Investigates Cases Step by Step

Clinical diagnosis often requires a sequence of decisions. A doctor gathers a patient's history, performs an examination, considers possible causes, orders tests and revises the working diagnosis as new evidence becomes available. This process can be particularly challenging when symptoms are unusual or several diseases could explain the same findings.

The Microsoft research describes an AI system designed to approach diagnosis in stages rather than analysing all available information at once. An orchestration system coordinates simulated clinical experts, progressively collects relevant information and updates the diagnostic plan as the case develops.

In reported tests involving more than 300 unusual and difficult cases, the system was more than twice as accurate as general physicians who used textbooks and the internet to investigate the cases. Researchers also reported that its proposed diagnostic work-ups cost less than those designed by human doctors in the tested scenarios.

These results are promising but do not demonstrate that the system is ready to replace clinicians. Performance on selected cases may differ from performance in busy hospitals, where records can be incomplete, patients have multiple conditions and errors can have serious consequences. The researchers said they planned further testing in an emergency-department setting.

AI diagnostic systems should be evaluated for safety, bias, transparency, privacy and real-world clinical effectiveness before routine use. Patients should continue to seek qualified medical care rather than rely on an AI tool for urgent or complex symptoms.