One of the arguments made for adopting structured report templates is that they help radiologists from overlooking important elements, especially for reports that are complex. Dr. Bao Do, a clinical instructor at Stanford University Medical Center, and colleagues decided to determine whether an NLP tool could alert a radiologist to elements that should be included in a report but were missing at the conclusion of a dictation using a speech recognition system.
In their experiment, they developed and validated an NLP tool to extract core features in assessing bone lesions on an x-ray image. The tool was then used to process 47 unstructured reports and categorize omissions by semantic categories. The tool correctly categorized 94%.
What are they going to do next? Attend this session to find out.
