SR evangelist touts technology's benefits, drawbacks

VANCOUVER - Speech recognition (SR) has matured to the point where it is user-friendly for most people, as long as they maintain realistic expectations of the technology, according to a presentation Thursday at the Society for Computer Applications in Radiology (SCAR) meeting.

Dr. Stephen Herman, associate professor at the University of Toronto and a staff radiologist at the University Health Network, Mount Sinai Hospital in Toronto, offered his thoughts on the benefits of adopting SR -- such as improved report turnaround time and cost savings -- as well as the drawbacks, e.g. the increased time spent on editing reports.

"The success of a speech recognition system is based on a clear understanding of the reality of these systems and what they can and cannot do," he said.

Herman, who has used SR at his practice since 1994, noted two main benefits accruing from the technology: faster report creation and the need for fewer or no transcriptionists.

"The problem with standard report creation (dictation and transcriptionist) is that there are delays in the typing of the report, delays in accepting the report, and the possibility that the report will be created twice (a verbal wet read and a final report)," Herman said.

SR technology permits the creation of the report as the case is viewed. The final report can then be completed in as little time as a few minutes after a diagnostic read is finished.

"The benefit to this is that the radiologist has total and complete control over the report generation process," said Herman.

However, this total control comes at a price: SR users must assume editorial responsibilities for their creations, a task which radiologists may resist despite the fact that they already do it.

"The reality is that most of us who are using a conventional dictation and transcription model are already doing some editing with their transcription-generated reports," noted Herman.

Content creation within the report body must also be more carefully crafted. Speech dysfluencies, such as stammering or the use of pause sounds such as "uh" or "ah," need to be minimized or eliminated, he said. A benefit of the increased attention by the radiologist on the report creation process is shorter, more concise reports.

"Radiologists tend to shorten their report length when they're using SR," he said. "According to some reports, the average length can go down by 30% or more, from 90 words to 65 words or less."

Successful adoption of the technology also requires a champion -- someone in the department or practice whose opinion is respected by other personnel. SR technology is not a turnkey application that can be loaded and run, as each user must be trained on the system, and prompt support needs to be provided on an ongoing basis, Herman emphasized.

He suggested that users start their first production day on the system with a light clinical load as early frustration leads to ongoing dissatisfaction and less accuracy with the application.

"For example, if a radiologist would normally be expected to dictate 20 CTs in the morning, 10 or 15 of these should be distributed to other colleagues in the department, leaving only five to report on with the new system. This will allow the radiologists to take their time and not feel pressure to complete the studies," he said.

Herman also advocated that SR adopters maintain a clear and unequivocal path toward the technology for a successful implementation. When there is no dictation/transcription alternative available, users will dedicate themselves to making the system work.

"It's imperative that when the system is rolled out there is no chance of turning back," he said.

By Jonathan S. Batchelor staff writer
May 21, 2004

Related Reading

Speech recognition shows strength in teaching hospital, April 23, 2004

Imaging reports also benefit from improved signal-to-noise ratio, May 3, 2004

SCAR members offer solutions for data deluge, February 24, 2004

Machine language: Radiologist offers tips on optimizing speech recognition, February 4, 2004

Adding the human touch to speech recognition, August 8, 2003

Copyright © 2004

Page 1 of 603
Next Page