By Cynthia E. Keen, AuntMinnie.com staff writer
November 11, 2011
Thursday, December 1 | 12:15 p.m.-12:45 p.m. | LL-INS-TH7A | Lakeside Learning CenterThe subject of this poster presentation is how Stanford University researchers validated the performance of a natural language processor (NLP) that they developed to calculate radiologist performance statistics, and how this is evolving into another project that may directly benefit patients.
The NLP was developed to automatically calculate a radiologist's performance for a specific diagnosis using unstructured reports in a RIS. For this initial test, the researchers selected the diagnoses of osteoporosis and osteopenia. The NLP was trained using 300 dual-energy x-ray absorptiometry (DEXA) and pelvic x-ray reports so that it would identify reports assessing normal and abnormal bony mineralization.
Session presenter Dr. Bao Do, a clinical instructor and radiologist working in the musculoskeletal section of the department of radiology, will present findings from an analysis of 1,910 pelvic x-ray reports and 1,496 DEXA reports. The NLP correctly classified all but one of the reports.
"The NLP was able to review several hundred thousand reports in less than a day's time, with a more than 99.9% accuracy rate," Do told AuntMinnie.com. "In an analysis we conducted over a five-year period, we determined that a pelvic x-ray exam has a high positive predictive value [PPV] of over 80% in identifying osteoporosis. Imagine the scenario of a patient who may not have health insurance or the funds to have a DEXA test who presents to an emergency department for hip pain after a fall. If an x-ray has a high PPV for identifying osteoporosis, the treating physician might be able to refer this patient to appropriate follow-up."
"So, our question is, without increasing the cost of healthcare, can an existing study such as a routine, inexpensive x-ray exam be useful as a complimentary screening tool to help identify more patients at risk for osteoporotic complications?" he said.
Last Updated np 11/10/2011 12:03:49 PM