Microsoft develops AI system that diagnoses patients step by step like doctors

Microsoft’s new artificial intelligence model has delivered more than twice the diagnostic accuracy of general practitioners in complex medical cases. Researchers also plan to explore its potential for use in real clinical settings.

Microsoft has developed a new AI tool that assesses patients’ conditions step by step, much like doctors, and determines whether additional examinations are needed. According to Reuters, a study published in Science on October 22 reported that the system achieved strong results in diagnosing complex medical cases.

Traditional medical AI systems generally attempt to make a diagnosis after receiving a patient’s symptoms, laboratory results and medical imaging data. Microsoft’s new system takes a different approach.

The model breaks the diagnostic process into stages. First, it determines which questions should be asked about the patient’s medical history and physical examination. It then analyzes the information gathered, selects the necessary examinations and tests, evaluates the results and orders additional tests when needed.

The process also involves a specialized “orchestrator” that coordinates the AI’s activities. It organizes the sequence of diagnostic decisions, coordinates the work of AI components that simulate different clinical specialists and allows initial hypotheses to be updated as new information becomes available.

As part of the study, the system was tested on more than 300 unusual and diagnostically challenging medical cases. According to the researchers, the model was more than twice as accurate as general practitioners who used textbooks and online resources to diagnose these cases.

The system also outperformed human doctors in scenarios developed using data from patients presenting to emergency departments.

Researchers note that previous assessments of medical AI systems have often relied on multiple-choice tests. In real clinical practice, however, doctors are not given a ready-made list of symptoms or a neatly organized medical summary. Instead, they must gather the necessary information by asking questions and conducting examinations.

According to Dominic King, Microsoft’s corporate vice president of health at its AI division, artificial intelligence has become highly proficient at multiple-choice medical tests, but diagnosis in real life is a much more complex process.

The study also assessed not only the system’s diagnostic accuracy but also its economic efficiency. According to the researchers, conducting diagnostic assessments using the new approach could cost less than assessments performed by human doctors.

Harsha Nori, head of AI at Microsoft AI Health, emphasized that the evaluation focused not only on reaching the correct diagnosis but also on how efficiently and carefully the system carried out the diagnostic process.

In the next phase, researchers plan to test the system in collaboration with Beth Israel Deaconess Medical Center in Boston. One of the main objectives will be to assess the tool’s ability to conduct initial patient assessments and triage patients according to the urgency of their conditions in emergency department waiting rooms.

The study’s findings do not indicate that the system is already capable of replacing doctors independently. Further testing will be needed to establish the new technology’s safety, reliability and practical benefits in real clinical settings.