
IIIT-Hyderabad study warns against blind trust in medical AI models
The story in brief
A study by IIIT-Hyderabad researchers cautions against blindly trusting medical AI models, highlighting that they may highlight correct image areas but fail to identify diseases as radiologists would. The research team, led by Prof. Parameswari Krishnamurthy and Dr. Syed Faizan, audited four vision-language models on chest X-rays and conducted a reader study with two radiologists to assess anonymised overlays. The study, accepted at MICCAI 2026 in Strasbourg, suggests AI models may not provide the same diagnostic accuracy as human experts, prompting radiologists to reconsider their reliance on such tools.