RadGraph (RadGraph: Extracting Clinical Entities and Relations from Radiology Reports)

RadGraph 是基于我们新颖的信息提取模式的放射学报告中的实体和关系数据集，由 600 个带有 30K 放射科医生注释的报告和 221K 个带有 10.5M 自动生成注释的报告组成。我们发布了一个开发数据集，其中包含来自 MIMIC-CXR 数据集（14,579 个实体和 10,889 个关系）的 500 份放射学报告的董事会认证放射科医生注释，以及一个测试数据集，其中包含两组独立的董事会认证放射科医生注释，用于 100 个放射学报告在 MIMIC-CXR 和 CheXpert 数据集中平均分配。我们还发布了一个推理数据集，其中包含为 220,763 个 MIMIC-CXR 报告（约 600 万个实体和 400 万个关系）和 500 个 CheXpert 报告（13,783 个实体和 9,908 个关系）自动生成的注释，并映射到相关的胸片。

RadGraph is an entity and relation extraction dataset for radiology reports based on our novel information extraction schema. It consists of 600 reports with 30K radiologist annotations, and 221K reports with 10.5M automatically generated annotations. We have released a development dataset containing board-certified radiologist annotations for 500 radiology reports from the MIMIC-CXR dataset, which includes 14,579 entities and 10,889 relations. Additionally, we have a test dataset with two independent sets of board-certified radiologist annotations for 100 radiology reports evenly split between the MIMIC-CXR and CheXpert datasets. We have also published an inference dataset with automatically generated annotations mapped to their associated chest radiographs, covering 220,763 MIMIC-CXR reports (approximately 6 million entities and 4 million relations) and 500 CheXpert reports (13,783 entities and 9,908 relations).

搜集汇总

背景与挑战

背景概述

RadGraph 是一个用于从放射学报告中提取临床实体和关系的医疗自然语言处理数据集，包含600份人工标注报告和约22万份自动生成报告。它提供了开发、测试和推理数据集，数据来源于MIMIC-CXR和CheXpert，旨在支持实体关系联合抽取任务。