CorBenchX
收藏资源简介:
CorBenchX是一个大规模的胸部X光报告错误数据集,包含26,326个胸部X光报告,其中包括24,164个单一错误案例和2,180个多重错误案例。每个报告都被标注了错误跨度、错误类型和简要描述。数据集旨在为放射学报告中的错误检测和校正提供一个全面的基准,用于评估和提升人工智能辅助的质量控制。该数据集是在MIMIC-CXR数据集的基础上,通过DeepSeek-R1提示注入临床常见错误构建的,为开发放射学报告错误检测和校正系统提供了一个可靠的基础。
CorBenchX is a large-scale chest X-ray report error dataset containing 26,326 chest X-ray reports, including 24,164 single-error cases and 2,180 multiple-error cases. Each report is annotated with error span, error type and brief description. This dataset aims to provide a comprehensive benchmark for error detection and correction in radiology reports, to evaluate and enhance AI-assisted quality control. Constructed based on the MIMIC-CXR dataset by injecting clinically common errors via DeepSeek-R1 prompting, it offers a reliable foundation for developing radiology report error detection and correction systems.




