RadGraph Development Dataset

该数据集包含了来自500份胸部X光放射学报告的标注实体和关系，由持有执照的放射科医师进行标注。数据集被划分为训练集和开发集，确保两个集合之间没有患者重叠。规模上，该数据集包含500份报告，共有14,579个实体和10,889个关系。任务目标是从事放射学报告中提取实体和关系。This dataset contains annotated entities and relationships sourced from 500 chest X-ray radiology reports, which were annotated by board-certified radiologists. The dataset is split into training and development sets, with no overlapping patients between the two subsets. In terms of scale, the dataset comprises 500 reports, with a total of 14,579 entities and 10,889 relationships. The task objective is to extract entities and relationships from radiology reports.

搜集汇总

背景与挑战

背景概述

RadGraph Development Dataset是一个包含放射学报告中临床实体和关系标注的数据集，旨在支持NLP和多模态研究。它包括专业标注的500份训练报告和模型自动标注的220,763份报告，覆盖MIMIC-CXR和CheXpert数据集。