Washington: Microsoft researchers have developed an AI system that diagnoses medical conditions step by step, achieving 83 percent accuracy on complex cases, compared with 36 percent for general physicians using reference materials.
Published in Science, the study tested the system on 304 complex cases and emergency department cases. It reached 80 percent accuracy on emergency cases, compared with 47 percent for physicians.
The system simulates a panel of clinical experts, gathering medical history, requesting tests and updating diagnostic hypotheses as new information emerges. It also assesses testing costs to identify more efficient diagnostic pathways.
The researchers said the approach better reflects real-world diagnosis than conventional AI tests, which provide all patient information at once. However, the findings are based on simulated cases and require further evaluation in clinical practice.
