RSNA LLM Benchmark Dataset for Chest Radiographs of Cardiothoracic Disease (REVEAL-CXR)
收藏资源简介:
REVEAL-CXR是由北美放射学会联合全球10个机构的17位心胸放射专家构建的胸部X光诊断基准数据集,包含200例经三重专家验证的DICOM格式影像(100例公开),涵盖12种心胸异常标签如肺实变、气胸等。数据源自MIDRC注册的13,735例临床影像,通过GPT-4o和Phi-4-Reasoning模型辅助提取报告特征,并采用分层抽样确保罕见病与多病症案例的覆盖。该数据集旨在解决现有放射学数据缺乏专家直接图像标注的问题,为评估多模态大语言模型的临床诊断能力提供黄金标准,特别关注复杂病例和罕见病种的模型性能验证。
REVEAL-CXR is a chest X-ray diagnostic benchmark dataset developed by the Radiological Society of North America (RSNA) in collaboration with 17 cardiothoracic radiologists from 10 global institutions. It consists of 200 DICOM-format imaging studies triple-validated by experts, 100 of which are publicly available. The dataset encompasses 12 cardiothoracic abnormality annotation labels, including pulmonary consolidation, pneumothorax, and other thoracic pathologies. Derived from 13,735 clinical imaging cases registered in the MIDRC registry, report features were extracted with the assistance of GPT-4o and Phi-4-Reasoning models, and stratified sampling was employed to ensure adequate coverage of rare disease cases and multi-morbidity presentations. This dataset aims to address the critical gap that existing radiological datasets lack direct expert-provided image annotations, serving as a gold standard benchmark for evaluating the clinical diagnostic capabilities of multimodal large language models (LLMs), with a particular focus on validating model performance on complex clinical cases and rare disease entities.

- 1RSNA Large Language Model Benchmark Dataset for Chest Radiographs of Cardiothoracic Disease: Radiologist Evaluation and Validation Enhanced by AI Labels (REVEAL-CXR)威尔康奈尔医学院·放射学系; 托马斯杰斐逊大学·放射学系; 多伦多大学·圣迈克尔医院/Unity Health Toronto医学影像系; 加州大学旧金山分校·放射学与生物医学影像系; 俄亥俄州立大学韦克斯纳医学中心·放射学系; 克利夫兰诊所基金会·诊断研究所; 以色列医院Albert Einstein; 北美放射学会; 哈佛医学院·布里格姆妇女医院放射学系; 宾夕法尼亚大学佩雷尔曼医学院·放射学系; 慕尼黑工业大学·诊断与介入放射学系; 约翰霍普金斯大学医学院·放射学系; 多伦多大学医学院·三一健康伙伴医学影像系; 西安大略大学·圣约瑟夫医疗保健伦敦; 埃默里大学医学院·放射学与影像科学系; 德克萨斯大学西南医学中心·放射学系 · 2026年



