Microsoft researchers have created a new artificial intelligence tool that can consistently make medical diagnoses — just like experienced doctors do. According to a study, the new system was more than twice as accurate as general practitioners when solving complex clinical cases.
Microsoft researchers have created a new artificial intelligence tool that can consistently make medical diagnoses — just like experienced doctors do. According to a study, the new system was more than twice as accurate as general practitioners when solving complex clinical cases.
Most existing medical AI models analyze all of a patient's data at once: they simultaneously process symptoms, test results, and images to instantly produce a ready-made diagnosis.
Instead, Microsoft's new development uses a special "orchestrator." It simulates a council of virtual clinical experts and builds the diagnostic process in stages, Reuters reports .
First, the AI decides what questions to ask the patient about their medical history and what to look for during a physical examination. Based on the initial data, the model determines which laboratory or instrumental tests to order. Once the results are received, the AI evaluates them, orders additional tests if necessary, and constantly updates its working diagnostic hypotheses.
When tested on over 300 complex and confusing medical cases, the orchestrator model showed twice the accuracy in patient management and establishing a final diagnosis than therapists who had access to textbooks and the Internet.
The tool also outperformed human doctors when analyzing real-world scenarios from a database of emergency departments.
“AI has already learned to take 100% multiple-choice medical tests. But this has no relevance to real life. No patient has ever come to my appointment with a nice, ready-made summary of their illness,” said Dr. Dominic King, vice president of Microsoft AI’s medical division.
In addition to increasing accuracy, the researchers discovered another important fact: a diagnostic process planned by an “orchestrator” costs less than examinations ordered by live doctors.
“We measure not only the accuracy of the systems. We also evaluate how thoughtfully and carefully the AI goes through the entire diagnostic examination,” emphasized Harsha Nori, head of the AI direction at Microsoft AI Health.
In collaboration with Beth Israel Deaconess Medical Center in Boston, researchers plan to test the new tool's ability to perform initial triage of patients directly in hospital emergency rooms.
