CheXthought: Global dataset of clinical reasoning for chest X-ray AI

CheXthought is a global multimodal dataset containing 103,592 chain-of-thought reasoning traces and 6,609,082 visual attention annotations from 501 radiologists across 71 countries, covering 50,312 multi-read chest X-rays. It captures cognitive processes and visual attention underlying clinical reasoning, unlike existing datasets that only pair images with reports. Analysis reveals distinct visual search strategies, integration of clinical context, and communication of uncertainty among experts. The dataset demonstrates clinical utility: CheXthought reasoning outperforms state-of-the-art vision-language model chain-of-thought in factual accuracy and spatial precision. This resource aims to improve AI interpretability and diagnostic reliability in chest X-ray interpretation.