Joseph Edwin Burns MD, PhD, Presenter: Nothing to Disclose

Ronald M. Summers MD, PhD, Abstract Co-Author: Royalties, iCAD, Inc Grant, iCAD, Inc Stockholder, Johnson & Johnson Grant, Viatronix, Inc

RIS search and report retrieval functions are valuable tools in teaching and research. However, the limited search query options of the RIS can result in a return of voluminous dictated report results for simple queries. Manually sorting these results can be a time intensive task for the radiologist. R is an open-source subroutine library for data analysis and graphics display. The purpose of this work is to develop and test a simple-to-use script set based on the R environment to tailor the RIS generated report set to the specific search characteristics of interest, streamlining radiologists’ research workflow.

RIS search was performed using the limiting criteria of Modality (CT) and Exam date (06/01/10-10/01/10). Free-text query criteria were: “ct+spine+fracture+cervical.” Results were saved into a file of concatenated dictated reports, utilizing the default RIS “Export to file” function. Each individual RIS report contained at least one instance of the word “fracture.” An R script was designed to read the RIS output free-text data file and separate the concatenated reports. The separated reports were then filtered and indexed, and individual report sections and sentences parsed. RIS derived reports with a sentence containing the word “fracture” without the word “no” were classified as true positive (TP). Reports containing the words “no” and “fracture” within a single sentence were classified as false positive (FP) cases. True positive reports were saved to a file by the R script.

Our RIS search identified reports from 514 cases. The RIS output file was then read into the R script program. The R output file contained 101 filtered reports. Manual review of the RIS and R script output files by a radiologist found 5 FP cases missed by the script. No TP cases were missed. The R script thus eliminated 98.8% of FP cases. For per-report analysis, the precision was 95% (95% CI:[ 88%, 98%]), with a recall of 100% (95% CI:[ 95%, 100%]).

This R environment script system can robustly filter free-text radiology reports via report and sentence segmentation, specifically eliminating reports in which the search term occurred in the context of negation in this test scenario.

A software system for search term filtering and indexing RIS search results was developed. This system provides a simple, flexible, and easy-to-use open source solution to radiology report data mining

Burns, J, Summers, R, Accelerating the Clinical Research Study Search: R Script Filtering of Dictated Reports from Radiology Information System (RIS) Searches.  Radiological Society of North America 2012 Scientific Assembly and Annual Meeting, November 25 - November 30, 2012 ,Chicago IL. http://archive.rsna.org/2012/12043654.html