Extraction, Labeling, Clustering, and Semantic Mapping of Segments From Clinical Notes

Area of Science:

• Natural Language Processing
• Clinical Informatics
• Computational Linguistics

Background:

• Limited tools exist for unsupervised information extraction from clinical notes in underrepresented languages like Czech.
• Accurate extraction is crucial for tasks like cancer registry reporting and patient record integration.

Purpose of the Study:

• To develop and evaluate a method for unsupervised extraction, classification, and clustering of semantically-labeled textual segments from clinical notes.
• To address the scarcity of computational tools for Czech clinical text analysis.
• To create a tool for computer-assisted semantic mapping to ontologies.

Main Methods:

• Unsupervised extraction of semantically-labeled textual segments from free-text clinical notes.
• Application of the method to a dataset of Czech breast cancer patients.
• Development of a tool for mapping extracted segment types to pre-defined ontologies.
• Validation on a downstream task of category-specific patient similarity.

Main Results:

• Demonstrated the practical relevance of the proposed approach for Czech clinical notes.
• Successfully extracted, classified, and clustered segments corresponding to specific clinical features.
• Validated the utility of the semantic mapping tool for patient similarity analysis.

Conclusions:

• The developed unsupervised method is effective for information extraction from Czech clinical notes.
• The approach facilitates the creation of sophisticated analytical pipelines for underrepresented languages.
• This work represents a stepping stone for broader applications in clinical data analysis and patient representation.