Dependency Parser-based Negation Detection in Clinical Narratives.

Area of Science:

• Clinical Informatics
• Natural Language Processing
• Medical Data Analysis

Background:

• Negation detection is vital for accurate clinical condition compilation and phenotype detection.
• The existing clinical Text Analysis and Knowledge Extraction System (cTAKES) uses a limited word-distance approach for negation.
• This method struggles with complex negation patterns in clinical documents.

Purpose of the Study:

• To evaluate if dependency structure from the cTAKES dependency parser can enhance negation detection performance.
• To compare dependency-based negation rules against the standard cTAKES negation annotator.

Main Methods:

• Manually compiled negation rules based on dependency paths were developed and tested.
• The study utilized the dependency parser within the cTAKES system.
• Negation rules were based on syntactic context rather than fixed word distance.

Main Results:

• Dependency-based negation rules demonstrated superior performance compared to the current cTAKES negation annotator.
• The syntactic context provided by dependency paths allows for more sophisticated negation scope identification.
• This approach overcomes limitations of fixed word-distance strategies.

Conclusions:

• Dependency parsing significantly improves the accuracy of negation detection in clinical text.
• The proposed dependency-based negation strategy is a more effective alternative for clinical natural language processing.
• Enhanced negation detection supports more reliable clinical data interpretation and phenotype extraction.