A hybrid system for temporal information extraction from clinical text.

Area of Science:

• Natural Language Processing (NLP)
• Clinical Informatics
• Biomedical Text Mining

Background:

• Accurate temporal information extraction from clinical notes is crucial for understanding patient histories and treatment timelines.
• Existing methods often struggle with the complexity and nuances of temporal expressions in medical records.

Purpose of the Study:

• To develop a comprehensive system for identifying events, temporal expressions, and their temporal relations in clinical text.
• To participate and evaluate system performance in the 2012 i2b2 clinical NLP challenge on temporal information extraction.

Main Methods:

• The system comprised three modules: event extraction, temporal expression extraction, and temporal relation (TLink) extraction.
• TLink extraction utilized three classifiers for relations between events and section times, within sentences, and across sentences.
• Performance was evaluated on a manually annotated dataset using micro-averaged Precision, Recall, and F-measure.

Main Results:

• The system achieved high rankings in the 2012 i2b2 challenge.
• An F-measure of 0.8659 was obtained for temporal expression extraction (ranked fourth).
• The system achieved first place in both the end-to-end TLink track (F-measure 0.6278) and the TLink-only track (F-measure 0.6932).

Conclusions:

• The developed system demonstrated strong performance in temporal information extraction from clinical text.
• The modular approach, particularly for TLink extraction, proved effective and competitive.
• Further refinements led to marginal improvements in TLink extraction performance.