What makes TIES unique

Negation Detection

Consider these two sentences:

  1. The tumor exhibits perineural invasion around large nerve bundles.
  2. There was no definite evidence of perineural invasion.

While “perineural invasion” is mentioned in both sentences, only the first case has perinerural invasion. Because identifying important features as present or absent is a common task in medicine, such pertinent negatives may represent the majority of mentions for some features. This can produce many false positives in typical EMR searches and require manual review and filtering.

TIES largely eliminates this problem. During the processing phase, TIES uses ConText – a negation detection module to determine whether the concept mentioned in the report is present or absent. It is smart enough to consider several different ways in which a report might indicate absence. For example, “no evidence of,” “not found,” and “no definite evidence” are all understood by TIES to mean that the concept is negated.

TIES also makes it possible to search directly for concepts that are negated, as opposed to those where there is no mention. For example, you could search for cases in which there is a specific mention there is no perineural invasion.

WW Chapman, D Chu, JN Dowling
ConText: An algorithm for identifying contextual features from clinical text
Proceedings of the Workshop on BioNLP 2007: Biological, Translational, and Clinical Language Processing