Detecting negation and scope in Chinese clinical notes using character and word embedding

Area of Science:

• Natural Language Processing
• Medical Informatics
• Computational Linguistics

Background:

• Negation detection in free-text clinical notes is crucial for accurate data indexing but remains challenging, especially in Chinese due to language processing complexities.
• Traditional rule-based negation detection methods have limitations in comprehensiveness and portability.
• Developing automated tools is essential for improving the efficiency and accuracy of clinical data analysis.

Purpose of the Study:

• To construct a large, annotated Chinese clinical notes corpus for negation detection research.
• To develop and evaluate a novel negation detection tool specifically for Chinese clinical text.
• To assess the performance of character and word embedding features in Chinese clinical natural language processing for negation identification.

Main Methods:

• A Chinese clinical corpus was created from admission and discharge summaries.
• Sequence labeling systems were developed for negation and scope detection.
• Features utilized included bag of characters, bag of words, character embedding, and word embedding, with an additional feature for nested scopes.

Main Results:

• High inter-annotator agreement (Kappa=0.79) was achieved during manual annotation.
• The developed negation detection system achieved 99.0% F-score, significantly surpassing rule-based methods (79% F-score).
• The best performing system, utilizing word embedding features, reported 99.0% precision and 99.1% recall for negation detection and 94.6% F-score for scope detection.

Conclusions:

• A state-of-the-art negation detection tool for Chinese clinical free-text notes has been developed.
• Word embedding features demonstrate significant effectiveness for identifying negations in clinical text.
• The proposed method successfully identifies nested scopes with multiple negations, enhancing clinical data processing.