Spatial Relation Extraction from Radiology Reports using Syntax-Aware Word Representations.

Area of Science:

• Medical Informatics
• Natural Language Processing
• Radiology

Background:

• Extracting spatial relations from medical texts like chest X-ray reports is crucial for clinical understanding.
• Previous methods for spatial role labeling often benefit from incorporating syntactic information.

Purpose of the Study:

• To investigate the task of spatial role labeling for extracting spatial relations from chest X-ray reports.
• To propose and evaluate syntax-enhanced word representations for improved extraction of radiology-specific spatial roles.

Main Methods:

• Utilized a bidirectional long short-term memory (Bi-LSTM) conditional random field (CRF) as a baseline model.
• Incorporated additional Bi-LSTMs to encode syntax based on dependency tree substructures.
• Empirically evaluated the contribution of syntax integration methods for spatial role extraction relative to a spatial indicator.

Main Results:

• Syntax embeddings significantly improved the baseline method's performance in spatial role labeling.
• Achieved average F1 score improvements of 1.3 for TRAJECTOR, 0.8 for LANDMARK, 4.6 for DIAGNOSIS, and 4.6 for HEDGE roles.

Conclusions:

• Syntax-enhanced word representations are effective for improving spatial role labeling in chest X-ray reports.
• The proposed method offers a promising advancement for automated analysis of radiological reports, enhancing the extraction of critical spatial information.