Natural language processing helps mine data on follow-up imaging exams

Tuesday, November 30 | 10:50 a.m.-11:00 a.m. | SSG08-03 | Room S402AB
In the era when radiology reports were printed on paper and automated intelligent data mining using natural language processing was just an idea, the thought of analyzing a million radiology reports for trends in procedure recommendation rates would have been preposterous.

But today it can be done. In this scientific presentation by researchers at Massachusetts General Hospital (MGH) in Boston, trends in recommendation rates from 1993 through 2009 for high-cost versus low-cost exams following a primary abdominal ultrasound study are reported.

The research team used its natural language processing program (Leximer, Nuance Communications, Burlington, MA), first to identify all reports of abdominal ultrasound exams. Out of the more than 1 million reports identified, a follow-up exam was recommended for approximately 20%.

An additional ultrasound exam was recommended in 14% of these cases, followed by recommendations for a biopsy (11%), an MRI exam (8%), and a CT exam (6%).

Supriya Gupta, MD, clinical research coordinator in the department of abdominal imaging at MGH, will discuss the team's findings, including the trend that recommendations for MRI as a follow-up exam increased from 4.55% in 1993 to 10.11% in 2009. Recommendations for CT exams fell dramatically, from 16.3% in 1993 to 3.75% in 2009.

Gupta attributes this to the likelihood of higher test specificity when an MRI exam is performed as compared with a CT exam.

Page 1 of 603
Next Page