A dataset of chest X-ray reports annotated with Spatial Role Labeling annotations

Area of Science:

• Medical Informatics
• Natural Language Processing
• Radiology

Background:

• Radiology reports contain crucial spatial information about findings and diagnoses.
• Extracting this spatial information manually is time-consuming and prone to errors.
• Automated methods are needed to efficiently process large volumes of radiology reports.

Purpose of the Study:

• To present a new dataset of 2000 chest X-ray reports annotated with spatial information.
• To facilitate the development of automated spatial information extraction techniques for radiology reports.
• To enable advanced clinical applications leveraging structured spatial data from medical imaging reports.

Main Methods:

• Development of a dataset using Spatial Role Labeling on 2000 chest X-ray reports.
• Annotation of radiographic findings, anatomical locations, diagnoses, and hedging phrases.
• Identification of spatial expressions (Spatial Indicators) and their associated spatial roles (Trajector, Landmark, Diagnosis, Hedge).
• Utilizing the dataset to develop deep learning models for automatic extraction of Spatial Indicators and roles.

Main Results:

• A dataset with 1962 Spatial Indicators, 2293 Trajectors, 2167 Landmarks, 455 Diagnosis, and 388 Hedges was created.
• The dataset enables the training of deep learning models for spatial information extraction.
• Demonstrated the feasibility of automatic extraction of spatial indicators and roles from radiology reports.

Conclusions:

• The presented annotated dataset is a valuable resource for advancing research in medical natural language processing.
• Automated spatial information extraction from radiology reports can significantly enhance clinical applications.
• The developed deep learning methods show promise for efficient and accurate information retrieval from clinical text.