Natural language processing (NLP) has multiple applications to radiology but is underused in the field, according to a recent article in the Journal of the American College of Radiology. However, the earlier technology on which NLP is based--voice recognition software--still has yet to be accepted by many radiologists, a Diagnostic Imaging survey found.
According to the JACR study, NLP currently has three main applications in radiology:
- To flag patient records to support outcomes research;
- To pinpoint specific data points, such as individual imaging findings, for analysis and quality improvements;
- To help radiologists improve their documentation by creating reports that highlight key points.
George Hripcsak, a biomedical informatics professor at Columbia University, told Diagnostic Imaging that NLP could have these additional benefits:
- It can be used to search patient databases for similar findings, which helps residents practice their diagnostic skills;
- Some types of NLP can help care teams identify instances where suspicious findings have been overlooked;
- It can convert radiology reports into language that's easier for laypeople to understand.
However, the physician survey found that only half of radiologists liked their speech recognition software. Thirty percent were unhappy with it, and the rest apparently didn't respond or had no opinion.
NLP software is considered the next generation of voice recognition programs. If it is no more accurate in "understanding" text than voice recognition is in recognizing speech, however, physician trust will be a barrier to acceptance.