Natural Language Processing Underutilized in Radiology Despite Advanced Capabilities




Natural language processing, considered the next generation of voice recognition software, makes it easier for you to summarize, find, and retrieve data from radiology reports. But a recent study shows many of you still aren’t using it.

Nearly 50 years ago, speech recognition software debuted on the healthcare scene, and providers used it to record radiology report findings. Technology improvements have taken the software to the next level with natural language processing (NLP), and it now plays a significant role in quality improvement efforts, said Ronilda Lacson, MD, a radiology research associate at Brigham & Women’s Hospital. NLP takes the voice-created narratives and makes them structured and searchable.

“NLP makes sure physicians report findings appropriately,” Lacson said. “They can record information in such a concise form so that when patient histories are pulled for review they’re like a thin cut of focused data.”

In a study published in the September Journal of the American College of Radiology, Lacson and her colleagues identified three main uses for NLP. The software can pull records that meet specific criteria to support effective outcomes research. Various versions also let you pinpoint specific data points, such as individual imaging findings, for analysis and quality improvements. However, the most valuable, long-term NLP use, Lacson said, is the brief reports it can create to highlight key content and critical findings. Other radiologists can study these summaries to improve their future documentation.

Lacson said the technology is underused, but her study didn’t include utilization rates in the imaging industry. According to Lacson’s research, there are roadblocks to efficiently implementing NLP, and a recent non-scientific poll of Diagnostic Imaging readers found that, as an industry, these difficulties have you divided on whether you use or like it. Based on 145 responses, roughly 50 percent of you are pleased with voice recognition software. However, nearly 30 percent of you dislike it.

These barriers come from a lack of information, said George Hripcsak, MD, a biomedical informatics professor at Columbia University. For much of his career, Hripcsak has studied how to use NLP to support clinical research and patient safety efforts, and he said there are many challenges to widespread implementation.

“Many radiologists just don’t know what programs are out there or what they can do with them,” he said. “Not only that, but the radiology market is also small. It likely doesn’t attract a lot of attention from companies looking to sell NLP systems.”

In addition, Lacson pointed to the steep learning curve associated with NLP technology and the lack of standards in place for measuring the usefulness of the software as hurdles to overcome.
Even with all these obstacles, Hripcsak said NLP offers many opportunities to enhance medical education, as well as patient safety. You can use NLP to search patient databases for groups of records that share specific findings, he said. This teaching tactic exposes your residents to many cases with similar characteristics and gives them the opportunity to practice their diagnostic skills.

Some NLP versions can help providers work as a team to catch instances where suspicious findings have been overlooked. In these cases, NLP sends up a red flag if there hasn’t been any follow up on anything troubling that was identified in an imaging test and noted in a patient’s record.

In the age of healthcare portals that give patients immediate access to their medical records, NLP can be a translation tool for people who don’t have medical training, Hripcsak said.
“Many people have fairly low health literacy,” he said. “And, it’s important they understand what a radiologist says about their MRI or CT scan. NLP can put a radiologist’s report into easy-to-understand lay language.”

This article was originally posted at  http://ping.fm/Ipy4Z