Microsoft AI Diagnoses Patients Step-by-Step, More Accurate
Microsoft has developed an artificial intelligence tool that diagnoses patients step by step, similar to how doctors do it, which has proven more than twice as accurate as general practitioners in complex medical cases, according to a study published in Science on Thursday.
The AI model identifies and coordinates the questions that should be asked of patients, as well as the examinations and tests that need to be scheduled, using what the researchers term an “orchestrator,” Reuters reports. The system organizes how the artificial intelligence will handle a given problem by employing a simulated group of clinical experts who collaborate to solve diagnostic cases, according to the report.
Unlike existing medical AI models that typically use an “all at once” approach, reviewing all patient symptoms, lab results, and imaging at once to make a diagnosis, the orchestrator-based model first determines what to ask about the patient’s medical history and what is important during a physical examination.
Based on the information gathered, it then decides which tests to order, evaluates the results, and prescribes additional examinations if needed. The researchers continuously update their hypotheses and plans throughout the process.
When tested on over 300 extremely unusual and confusing medical dilemmas, the model using the diagnostic orchestrator was more than two times more accurate than general practitioners who rely on textbooks and the internet, according to the study.
Read Also: Spotify’s Growth Slows as Revenue Misses Targets
Evaluations comparing this type of older system to human doctors typically use multiple-choice questions, they explained.
“Artificial intelligence is already achieving excellent results in these tests,” said Dr. Dominic King, corporate vice president of health at Microsoft AI, whose team developed the new technology. “But of course, this is not the real world. No patient has ever come to me with a well-prepared summary.”
The diagnostic processes also cost less when the orchestrator plans the examinations, compared to when doctors conduct them, the researchers found.
“We are not just measuring the accuracy of systems. We are also measuring how rational and careful the diagnostic process is,” said Harsha Nori, head of the artificial intelligence department at Microsoft AI Health.
In collaboration with the Beth Israel Deaconess Medical Center in Boston, the researchers will test the next step, the tool’s ability to sort patients in emergency department waiting rooms.
