Community-based study establishes superiority of voice recognition software

November 25, 2007

A study from the Midwest that tracked the effects of converting from traditional transcription to voice recognition software indicates the automated approach is more accurate. The new approach also contributed to an impressive improvement in report turnaround time compared with manual report preparation.

A study from the Midwest that tracked the effects of converting from traditional transcription to voice recognition software indicates the automated approach is more accurate. The new approach also contributed to an impressive improvement in report turnaround time compared with manual report preparation.

Radiology Consultants of Iowa, a nonacademic radiology group in Cedar Rapids, IA, will generate about 400,000 dictations this year using voice recognition transcription software for two large acute-care hospitals, seven rural hospitals, and an imaging center. John Floyd, a partner in the 24-member group, described his findings Sunday at the 2007 RSNA annual meeting.

More than 97% of RCI's reports are self-edited and signed by radiologists immediately upon dictation, though they have the option to send their dictation to a correctionist at any time, Floyd said.

The study established error rates for 498 reports generated with voice recognition software and compared them with error rates for reports from the same imaging studies transcribed with traditional methods. Twenty-four radiologists each interpreted 20 to 25 studies involving CT, MR, and general radiographic procedures.

Floyd found that automated speech recognition was more accurate than traditional transcription. Nine percent of reports prepared with speech recognition included at least one transcription error, compared with 13% for traditional transcription. The rate for significant errors, requiring the preparation of an addendum, was 0.6% for speech recognition and 2% for traditional transcription, he said.

The accuracy rate for speech recognition at one of the two participating hospitals was confirmed by the facility's independent analysis. Its evaluation of 514 reports in September 2007 produced an overall transcription error rate of 9.7% for both approaches. An addendum was required for 0.6% of the reports.

Radiology report turnaround time improved at all sites. With traditional transcription, RCI finalized from 4% to 8% of its reports in less than 60 minutes in the two years before it adopted voice recognition in January 2007. The rate jumped to 66% during the group's first three months of experience with voice recognition. It then rose to 82% during the second three-month period.