Abstract Archives of the RSNA, 2013
Amilcare Gentili MD, Presenter: Nothing to Disclose
Brian E. Chapman PhD, Abstract Co-Author: Nothing to Disclose
Using pyContex is possible to correctly classify reports containing description of AAAs, simplifying the task of finding patients that may need a follow-up.
Current recommendations are for one-time screening for abdominal aortic aneurysm (AAA) by ultrasonography in men aged 65 to 75 who have ever smoked, followed by yearly ultrasonographic screening if aortic diameter is between 3.0 to 4.0 cm; ultrasonography every 6 months if aortic diameter is between 4.0 to 4.5 cm; and referral to a vascular specialist if aortic diameter is greater than 4.5 cm. To separate radiology reports describing AAAs from reports describing normal aortas, we use natural language processing.
For this study, we used pyConText, a Python implementation of the ConText. ConText is a simple text-processing algorithm that uses simple lexical cues to relate modifying phrases, such as expressions of uncertainty, temporality, or negation, to findings described in text. The classification performed by a radiologist reviewing the radiology reports was compared with pyConText classification.
Out of 473 reports pyConText classified 82 patients as having an AAA, and 391 as not having an AAA including 4 false negative and 5 false positive for a sensitivity of 95.1% and a specificity of 98.7%.
Gentili, A,
Chapman, B,
Use of Natural Language Processing to Classify Radiology Reports Containing Description of the Abdominal Aorta. Radiological Society of North America 2013 Scientific Assembly and Annual Meeting, December 1 - December 6, 2013 ,Chicago IL.
http://archive.rsna.org/2013/13025753.html