For several years, researchers at MD Anderson Cancer Center in Houston have been developing ViSion, a multimedia structured reporting software application. ViSion captures key images and voice descriptions of imaging findings, tags the images with metadata referenced to an ontology, and assembles a multimedia structured report with historical data linked in timelines, said presenter Dr. David Vining. When the first prototype was developed in 2011, the software used pull-down menus to label findings and check-box menus for describing additional structured detail for each finding.
The researchers knew, however, that this slow and tedious manual data-entry process would never succeed in a busy clinical practice. As a result, they integrated a natural language processing (NLP) engine into ViSion to automate the labeling of findings and input of additional structured detail, he said.
"It was definitely a eureka moment when we first got it to work last year, and it works quite well," Vining said. "We are still refining the NLP performance and incorporating synonyms and diagnostic phrases because radiologists have many ways of saying the same thing."
Indeed, structured reporting with the production of discrete structured data is possible using NLP, but some challenges still need to be solved, he said. For example, all synonyms and phrases need to be incorporated into an ontology -- a controlled vocabulary with defined relationships between terms.
"However, with the integration of NLP into a structured reporting process, structured reporting has a real chance of succeeding in clinical practice and improving the quality of report data that a radiologist produces," Vining told AuntMinnie.com. "And that, in turn, will benefit big-data analytics and applications."
