A classification approach to coreference in discharge summaries: 2011 i2b2 challenge

Area of Science:

• Natural Language Processing
• Medical Informatics
• Computational Linguistics

Background:

• Accurate coreference resolution is crucial for understanding clinical narratives in electronic health records.
• Discharge summaries contain complex linguistic structures requiring sophisticated natural language processing (NLP) techniques.
• The 2011 i2b2 challenge focused on coreference resolution within clinical text.

Purpose of the Study:

• To develop a high-performance coreference resolution system for medical discharge summaries.
• To address the 2011 i2b2 coreference challenge, focusing on Person, Problem, Treatment, and Test entities.
• To improve the extraction and understanding of key information from clinical notes.

Main Methods:

• An integrated coreference resolution system was designed, combining three subsystems.
• The system leveraged Person attributes, contextual semantic clues, and external world knowledge (e.g., Wikipedia).
• Subsystems included Person coreference, Problem/Treatment/Test coreference, and Pronoun resolution using a support vector machine classifier.

Main Results:

• The system achieved a top ranking in the 2011 i2b2 challenge, outperforming 20 competing teams.
• Overall micro-averaged F-measure of 0.915 was obtained on the challenge's test data.
• The highest performance was observed in the Person coreference sub-task.

Conclusions:

• The developed system demonstrates high accuracy and feasibility for coreference resolution in discharge summaries.
• The integration of world knowledge and contextual semantic extractors significantly enhanced the Problem/Treatment/Test system.
• This work contributes to advancing NLP applications in clinical data analysis and information extraction.