AP胸X光片句子级发现基准数据集

本研究介绍了由IBM Almaden研究中心创建的‘AP胸X光片句子级发现基准数据集’，该数据集包含73个丰富的句子级描述，用于描述AP胸X光片中观察到的发现。数据集通过半自动化的真值生成过程，利用众包方式收集临床医生的注释来获取这些发现。该数据集旨在通过高粒度的标签训练分类工具，解决现有数据集标签不足的问题，并特别关注AP视图中胸X光片的发现，如中央血管线和管道的放置。此数据集的应用领域包括自动化报告生成和辅助放射科医生进行诊断。

This study presents the 'AP Chest X-Ray Sentence-level Findings Benchmark Dataset' created by the IBM Almaden Research Center. This dataset contains 73 detailed sentence-level descriptions for characterizing observed findings in AP chest X-rays. The dataset is constructed via a semi-automated ground-truth generation pipeline that collects clinician annotations through crowdsourcing to obtain these findings. It aims to train classification models using fine-grained labels to address the issue of insufficient labeled data in existing datasets, with a particular focus on findings in AP-view chest X-rays, such as central vascular line and tube placement. Potential application scenarios of this dataset include automated radiology report generation and assisting radiologists in their diagnostic work.