Computer-digitized speech that can be accurately displayed and printed has been a promise waiting to happen for radiologists trying to close the gap between viewing an image and forwarding a report to a referring physician.

“The pressure is tremendous,” said MUSC radiologist Kenneth Spicer, M.D., Ph.D., explaining that a long-enough delay could mean loss of third party payment on an imaging procedure.

“Our job is diagnosing problems in patients,” Spicer said. “We look at images and interpret them. We dictate our impressions, which are then written in a report and forwarded to the referring physician who acts upon our findings.”

The time crunch, Spicer said, comes in the transcription process where information dictated by the radiologist is typed into a computer by stenographers, printed, proofed for errors, corrected and reprinted in its final form. As the dictations pile up when the transcription process backs up, the radiologist is facing a one- to two-day wait that delays medical care and threatens insurance coverage.

So, time really is money, and that factor makes it cost effective to adopt a dependable, accurate voice recognition system capable of immediately returning a nearly error-free document from spoken words to the radiologist for final signature.

“We struggled for five or six years with the first voice recognition product,” Spicer said, “but just recently we’ve finally found what we need.”

First there was the Kurtzweil software that required speaking into a computer and watching the words appear on a screen. After all the hype, its difficulty with medical terms and accents made it impractical to use.

Then IBM’s Medspeak, a product with a much broader vocabulary that could handle accents once it was trained to hear them, was tried. “It didn’t perform well in our setting,” Spicer said. He explained that as words appeared on the screen the radiologist would have to stop and make corrections. “It just didn’t emulate what a radiologist does.” The latest initiative appears to be a winner for radiologists, at least.

Spicer said a voice recognition system called Powerscribe, which was developed at Emory University and uses VR software by Dragon offers the radiologist tremendous flexibility with 95 percent accuracy the first time it’s used and improves to more than 99 percent with 30 minutes of training.

Loaded with the typical terminology used by radiologists, Powerscribe is a dictation system that does not require its user to sit at a computer. Instead, the radiologist uses the equipment much the same as a dictation machine would be used. Spoken words are digitized and can be printed to near-perfect accuracy immediately.

“Reports that are dictated in 15 to 20 minutes are printed two to three minutes later,” Spicer said. He allowed that words unique to the system have to be taught to Powerscribe, and users have to develop consistency in dictating. That means being tired or in a hurry can affect the accuracy of the dictation, if the user lets it.

Spicer said a test run at MUSC in April was successful. By the second day, many radiologists in the department were asking to have a go at it. “It generated some excitement in the department,” he said. “It’s definitely something we need.”

VR in pathology

With hands gloved and wielding tools to probe and slice into diseased organs, the pathologist can hardly turn from the wet specimen tray to the computer keyboard to type a report. But type a report is exactly what has to be done, so the information is dictated to a recorder which is then typed by a medical transcriptionist.

“That means waiting until the next day to receive the transcript and then editing it to be sure it’s correct,” said David Lewin, M.D., an assistant professor in the Department of Pathology and Laboratory Medicine.

About eight months ago his department purchased a Kurtzweil voice recognition system that Lewin says serves the department well when used in its template format. As a specimen is examined, specific items are filled-in by one-word responses from a word bank of possible choices.

“When we’re examining specimens or biopsies, our hands need to be free,” Lewin said. “But still, we have to describe what we see. Traditionally we’ve dictated to a microphone.” He explained that most pathology procedures lend themselves to Kurtzweil’s template program which is “essentially fill-in-the-blank.”

Lewin said that during the past six months while the department has been using the VR system, it has been working reasonably well— about 90 percent accuracy with a 70,000-word vocabulary—in the template format, but not working as well with free speech.

“I think it’s better, because you’re doing it once and it’s done,” Lewin said. “Also, it standardizes all the reports.”