JACR: NLP remains underused in radiology
Although natural language processing (NLP) offers multiple useful applications, radiologists have not yet embraced the technology, according to an article published in the September issue of Journal of American College of Radiology.

Ramin Khorasani, MD, MPH, and Ronilda Lacson, MD, PhD, from Brigham and Women’s Hospital in Boston, reviewed NLP applications and offered a summary of barriers to adoption.

Three primary NLP applications are:
  • Information retrieval—focuses on pulling entire documents or reports that meet specific criteria. Binary classification, “disease present” or “disease absent,” enables automated cohort generation for outcomes research, patient identification for quality metrics or case identification.
  • Information extraction—recognizes and extracts specific entities such as imaging findings or recommended management into structured forms and can be applied for quality improvement activities.
  • Summarization—generates a brief version of the report that preserves key content. The automated NLP system could scan the report for critical findings to provide feedback to radiologists to improve documentation, wrote Khorasani and Lacson.

Despite the potential of NLP in radiology, it is underutilized, the authors said. They pointed to several barriers, including a steep learning curve with NLP, the absence of an annotated textual gold standard and lack of defined metrics for measuring usability.

Khorasani and Lacson painted an optimistic picture of the future of NLP and concluded, “There remains an immense potential for growth in this field, because understanding NLP and using applications that currently exist, let alone development of more advanced applications, can have a significant impact on streamlining workflow and improving healthcare quality.”