Extraction of Temporal Information from Clinical Narratives.

Area of Science:

• Natural Language Processing
• Clinical Informatics
• Biomedical Data Science

Background:

• Electronic Medical Records (EMRs) contain vast clinical text crucial for healthcare and research.
• Extracting temporal information (expressions, events, relations) from clinical text is complex due to implicit language and domain specificity.
• Existing methods relying on annotated corpora are costly and limited by data size.

Purpose of the Study:

• To develop a novel method for effective temporal information extraction from EMR clinical texts.
• To address the challenges of implicit temporal expressions and limited annotated data in clinical text processing.
• To improve the accuracy and efficiency of extracting temporal events and relations from EMRs.

Main Methods:

• Developed a feature set tailored for clinical expressions.
• Implemented a semi-supervised framework for temporal event extraction.
• Utilized a newly formulated hypothesis for detecting temporal relations among events.

Main Results:

• Achieved an F-measure of 89.98% for temporal event extraction.
• Achieved an F-measure of 67.1% for temporal relation extraction.
• Demonstrated improved performance compared to existing methods on the I2B2 dataset.

Conclusions:

• The proposed method effectively extracts temporal information from clinical EMR texts.
• The semi-supervised framework and novel hypothesis offer a promising approach for temporal information extraction.
• This work contributes to advancing clinical text processing and information extraction in healthcare.