An end-to-end system to identify temporal relation in discharge summaries: 2012 i2b2 challenge

Abstract

The system achieved encouraging results, demonstrating the feasibility of the tasks defined by the i2b2 organizers. The experiment result demonstrates that both global and local information is useful in the 2012 challenge.

Representations of Time Expressions for Temporal Relation Extraction with Convolutional Neural Networks

“…The CNN token-based model had similar performance as some of the one-tag temporal representations (Table 1( 3,4,5)). Removing the time expression entirely (Table 1(10)) did not hurt performance much, confirming that time expressions were not critical cues for within-sentence event-event relation reasoning (Xu et al, 2013). Thus, on the colon test set, we evaluated the contribution of encoding time expressions on the eventtime CNN model only.…”

Extracting Time Expressions from Clinical Text

“…The backwards BIO tagger achieved an Fscore of 0.889 on the i2b2 Challenge data allowing for partial matches (the Overlapping column). The best performing system in the i2b2 Challenge (Xu et al, 2013) is shown in the last row, with an F1 score of 0.914, with an advantage on recall. Our best system performance would tie for 4th in the span matching part of that challenge, without tuning for that dataset.…”

“…Our best system performance would tie for 4th in the span matching part of that challenge, without tuning for that dataset. While we incorporated features based on the best-performing similar system (Xu et al, 2013), including punctuation information, prepositions, and chunk information, these did not improve performance. Their paper described a larger system and did not contain enough detail on time expression extraction to replicate exactly (e.g., the Other Keywords section in the online supplement is not exhaustive and probably is important to their result).…”

“…In that task, a variety of approaches were used for time expression extraction. The best performing system (Xu et al, 2013) used machine learning, with a conditional random field classifier (CRF) for finding spans and a support vector machine classifier for classifying attributes. Other top approaches used adapted regular expressions (Sohn et al, 2013) on top of the off the shelf Heideltime system (a general-domain NLP system for parsing time expressions) (Strötgen and Gertz, 2010).…”

A Problem-Action Relation Extraction Based on Causality Patterns of Clinical Events in Discharge Summaries

“…The challenges include extracting clinical events and relation classification, and identifying temporal relation between two events. Many research groups have participated and proposed technologies of natural language processing for medical texts [1,2,3,4].…”