Microsoft develops AI tool that can diagnose ‘sequentially’ like a doctor

Microsoft researchers have developed an AI tool capable of diagnosing patients in a step-by-step manner, mirroring the approach of a doctor, according to a study published in Science. This model was found to be more than twice as accurate as general physicians in resolving complex medical cases, as reported by Nancy Lapid.

The AI tool, utilizing an "orchestrator," determines the questions to ask patients, the necessary exams and tests to order, and organizes the diagnostic process. Unlike existing medical AI models that typically analyze all symptoms, lab results, and imaging studies simultaneously, this tool first assesses the patient's history and physical exam, then decides on relevant tests, and updates its hypotheses and plans accordingly.

When tasked with diagnosing over 300 unusual and perplexing medical cases, the AI model using the diagnostic orchestrator demonstrated a significantly higher accuracy rate (over 100%) compared to human physicians relying on textbooks and the internet. The tool also outperformed human doctors in scenarios derived from a major database of emergency room admissions.

According to Dr. Dominic King, corporate vice president of the health unit at Microsoft AI, AI has become adept at achieving near-perfect accuracy on multiple-choice questions, which doesn't reflect real-world medical situations. The researchers noted that the AI's diagnostic processes were also more cost-effective than those performed by human doctors.

In a collaboration with Beth Israel Deaconess Medical Center, the researchers will next evaluate the tool's capability to triage patients in emergency department waiting rooms.