Trending
Microsoft researchers say a new artificial intelligence (AI) tool that diagnoses patients step by step outperformed general physicians in complex medical cases, according to a study published Thursday in Science.
The system was more than twice as accurate as general physicians at diagnosing highly unusual and challenging cases, the researchers reported. Unlike conventional medical AI systems, it follows a diagnostic process similar to the approach doctors use when assessing patients.
The technology determines which questions to ask, identifies relevant findings from physical examinations and selects appropriate tests. It then analyses the results and updates its diagnostic hypotheses as new information becomes available.
Many existing medical AI systems analyse a patient’s symptoms, laboratory results and medical images together before producing a diagnosis.
Nvidia CEO Jensen Huang Dismisses AI Extinction Warnings as “Cheap” Headlines
Microsoft’s new approach uses an “orchestrator” to coordinate the diagnostic process. The system employs a simulated panel of clinical experts to work through cases and determine the next steps.
First, it considers the patient’s medical history and relevant examination findings. It then decides which tests to order based on the information gathered.
As results arrive, the system reassesses its working diagnoses and determines whether further testing is necessary. This process allows the AI to refine its conclusions rather than relying solely on an initial assessment.
Researchers evaluated the system on more than 300 unusual and difficult medical cases. According to the study, the AI tool was more than twice as accurate as general physicians who used textbooks and the internet to investigate the cases.
The system also outperformed human physicians when researchers tested it on scenarios drawn from a major database of emergency department admissions.
The researchers said conventional evaluations of medical AI often rely on multiple-choice questions. Such tests can measure medical knowledge but may not reflect the uncertainty and complexity doctors face when real patients seek treatment.
“AI has become very good at getting a 100% pass rate on those,” said Dr Dominic King, corporate vice president of Microsoft’s health unit.
“But of course, that’s not real-world. No patient ever came to see me with a nice summary,” he added.
King’s team is developing the technology at Microsoft AI.
The study also found that diagnostic work-ups designed by the AI orchestrator cost less than those conducted by human doctors.
Harsha Nori, head of AI at Microsoft AI Health, said the research evaluates not only diagnostic accuracy but also how carefully and efficiently the system investigates a patient’s condition.
“We’re not just measuring the accuracy of the systems. We’re also measuring how judiciously and carefully it goes through a diagnostic work-up,” Nori said.
The findings suggest that a step-by-step AI approach could help improve diagnostic efficiency. However, further research will be needed to establish how well the system performs in routine clinical settings.
Microsoft researchers are preparing to evaluate the tool in collaboration with Beth Israel Deaconess Medical Centre in Boston.
The next phase will examine whether the technology can help triage patients in emergency department waiting rooms. Triage involves assessing patients’ conditions and determining the urgency of their treatment.
This testing could provide further evidence about the system’s practical value in healthcare settings.
Although the initial findings are promising, performance on challenging cases does not by itself establish that an AI system is safe or effective for unsupervised patient care. Its real-world performance, clinical limitations and ability to support doctors will require further evaluation.
