IIIT-H study cautions doctors against relying solely on AI for chest X-ray diagnosis

A study by researchers at the International Institute of Information Technology, Hyderabad (IIIT-H) warns that AI tools used to interpret chest X-rays may highlight regions that do not align with those identified by radiologists, urging doctors not to rely on them entirely for diagnosis. The study, led by the Language Technologies Research Centre, evaluated four vision-language models against thousands of chest X-rays with two radiologists participating. It found that AI heatmaps may be influenced by a diagnosis's anatomical expectations rather than independent image analysis, and that model performance differed between technical audits and clinical assessments. The research, accepted for presentation at MICCAI 2026, addresses a gap in medical AI evaluation focused on prediction accuracy rather than region matching. The team is also studying model sensitivity to question phrasing, emphasizing that AI should support clinicians, not replace them.