Paras Lakhani MD, Presenter: Nothing to Disclose

Christopher Geordie Roth MD, Abstract Co-Author: Author, Reed Elsevier

Richard Earnest Sharpe MD, MBA, Abstract Co-Author: Nothing to Disclose

Kristen Elizabeth McClure MD, Abstract Co-Author: Nothing to Disclose

Paul James Read MD, Abstract Co-Author: Nothing to Disclose

George Paul Hobbs MD, Abstract Co-Author: Nothing to Disclose

Vijay Madan Rao MD, Abstract Co-Author: Nothing to Disclose

Adam Eugene Flanders MD, Abstract Co-Author: Nothing to Disclose

Structured reporting (SR) is felt to have many advantages over free-text reporting, including that it is preferred by clinicians, facilitates data-mining, business analytics, retrospective research, and quantitative imaging.  However, traditional SR reporting applications were found to be time-consuming by some radiologists, resulting in decreased productivity. Thus, the purpose of this study was to determine the feasibility a natural language processing (NLP) solution to automatically create standardized reports from free-text radiology dictations. Such a solution could provide the benefits of structured reporting with minimal loss in productivity.  In this exhibit, we demonstrate the ability of an NLP solution to transform free-text cardiothoracic CT interpretations into structured reports.

  A web-based computer programming application using NLP techniques was developed at our institution to transform free-text cardiothoracic CT interpretations into structured reports. Examples of the software in converting free-text to structured reports will be provided. In addition, users will be able to enter in their own free-text cardiothoracic CT dictations, and test the software's ability to structure their reports in real-time.

  This NLP solution re-organizes the report by placing text into anatomy-driven subheadings.  The goal of this is to improve the readability and consistency of the reports.  The application can transform reports in real-time during sign-off or retrospectively on a database of reports.  The software uses common web-based programming languages (PHP, Javascript, HTML) and can integrate with different reporting and radiology information systems.  Future efforts are underway to adapt the lexicon of the free-text report into those supported by RADLEX.

Natural language processing can automatically generate structured cardiothoracic CT radiology reports from free-text.  The organization and content of such reports can be customized for institutional or individual preferences. 

Lakhani, P, Roth, C, Sharpe, R, McClure, K, Read, P, Hobbs, G, Rao, V, Flanders, A, Automatic Creation of Structured Cardiothoracic Computed Tomography Reports Using Natural Language Processing.  Radiological Society of North America 2013 Scientific Assembly and Annual Meeting, December 1 - December 6, 2013 ,Chicago IL. http://archive.rsna.org/2013/13015813.html