Brought to you by
ECR 2018 / C-2810
Multimedia structured reporting employing natural language processing to improve the efficiency of data entry
Congress: ECR 2018
Poster No.: C-2810
Type: Scientific Exhibit
Keywords: Computer applications, RIS, PACS, Structured reporting, Computer Applications-General, Quality assurance, Workforce
Authors: D. J. Vining, A. Pitici, C. Popovici, A. Prisacariu, M. Kontak; Houston, TX/US

Aims and objectives

Structured reporting is essential for the creation of data that can be mined for healthcare outcomes. The primary reason cited for MD Anderson Cancer Center's failure to create an "Oncology Expert Advisor" using IBM's Watson was the lack of structured data contained in the electronic health record [1,2].


Structured reporting in diagnostic radiology has existed in various forms for decades. In 1976 Wheeler reported on the Johns Hopkins Radiology Reporting System that required users to point at touch screens to compile a narrative radiology report using key words and phrases [3]. 


Johnson evaluated another computerized system in 2009 in which users indicated that the system was inefficient, difficult to use, overly constraining, prevented desired content, and produced reports that were very different from free-text reports [4].


In 2011 Schwartz reported on the success of using structured templates to improve reporting, and simultaneously the RSNA has invested considerable time and resources to develop hundreds of templates that are available to the public [5, 6]. However, as pointed out at a user-group meeting held at the ECR in 2013, template reporting may not be the optimal solution to structured reporting due to variability in user-defined templates and descriptors [7].


A significant hindrance to many structured reporting approaches is the requirement that users select data from pull¬≠-down menus, check lists, or enter data manually. To improve the efficiency of structured data input, we have developed a hybrid solution that allows a radiologist to practice naturally (i.e., identifying image findings and dictating descriptions) while simultaneously generating structured data in the background using natural language processing to tag images with metadata referenced to the SNOMED CT ontology, and then assembling a multimedia structured report with related information from serial exams linked in timelines (figure 1) [8].

POSTER ACTIONS Add bookmark Contact presenter Send to a friend Download pdf
2 clicks for more privacy: On the first click the button will be activated and you can then share the poster with a second click.

This website uses cookies. Learn more