Microsoft AI diagnoses complex medical cases more than twice as accurately as doctors, study finds

Microsoft has developed an artificial intelligence (AI) system that was more than twice as accurate as general physicians at diagnosing complex medical cases in a study published in Science. Unlike many existing medical AI tools, which analyse symptoms and test results together, the system works through a case step by step, much like a doctor.

The findings could point to a new role for AI in healthcare, particularly when symptoms are confusing or a diagnosis requires several rounds of testing. However, the results come from diagnostic assessments, and further testing will be needed to establish how well the technology performs in real hospitals.

How Microsoft’s AI diagnoses patients differently

The system uses a component called a diagnostic “orchestrator” to organise the process. It coordinates a simulated panel of clinical experts and decides what information is needed at each stage. First, it identifies questions about the patient’s medical history and the findings that matter during a physical examination. It then decides which tests to order, evaluates the results and determines whether further investigations are necessary.

As new information becomes available, the system updates its possible diagnoses and plans its next steps.

This differs from the approach used by many existing medical AI models, which assess a collection of symptoms, laboratory results and scans to suggest a diagnosis in one go.

AI was tested on hundreds of difficult medical cases

According to the researchers, the orchestrator-based system was more than twice as accurate as general physicians when working through more than 300 unusual and challenging medical cases. The doctors were allowed to use textbooks and the internet.

The AI also outperformed physicians in scenarios based on a major database of emergency department admissions. Microsoft AI health executive Dominic King said conventional tests often rely on multiple-choice questions, where AI systems can perform extremely well. Real patients, however, rarely arrive with symptoms neatly summarised in a question. The researchers also reported that the AI-designed diagnostic work-ups cost less than those carried out by human doctors. They said the evaluation considered not only diagnostic accuracy but also how carefully and efficiently tests were selected.

Could Microsoft’s AI be used in emergency rooms?

The next step is to test the technology in a more practical setting. Microsoft researchers are collaborating with Beth Israel Deaconess Medical Center in Boston to assess whether the tool can help triage patients in emergency department waiting rooms.

Triage involves assessing patients to determine how urgently they need medical attention. If the system proves reliable, it could potentially help healthcare teams manage workloads and identify patients who need faster assessment. But the study does not establish that AI can replace doctors. Performance in structured diagnostic cases does not automatically translate into safe decisions for real patients, where information may be incomplete and mistakes can have serious consequences.

For now, the findings suggest that AI could become a useful support tool for clinicians, particularly when cases are difficult to diagnose. Whether it can deliver those benefits consistently in hospitals remains to be tested.