Voice recognition enters realm of mature technologies

Software that turns spoken words into text reduces the time from dictation to finished report, streamlines report generation, and increases productivity. The technology is transforming the way radiology reports are prepared, yet its impact has hardly been felt.

Although voice recognition products have entered the mainstream of radiology, appearing in large and small hospitals and outpatient imaging centers, only about 10% of the potential market has adopted them.

At the RSNA meeting, vendors will try to change that statistic with promises of productivity and cost savings. Most of their claims are valid, as this once-experimental, cutting-edge technology has matured.

VR products provide reasonable accuracy in translating speech into typed words, offer productivity tools that capitalize on RIS-provided information, and afford convenient access to the software, often via the Internet. They are present, however, in only about 900 U.S. hospitals and clinics, according to Dictaphone, one of the pioneers of traditional and VR-based transcription equipment.

Dictaphone pegs the overall U.S. market for both manual and VR transcription at about 4200 hospitals and 5000 imaging clinics, and it estimates that its VR software is operating in about 500 of them. This may be enough, however, to dispose radiologists toward VR and away from traditional transcription.

"I get the sense that speech recognition has crossed the bridge of acceptance, and adoption will grow," said Don Falatti, senior vice president of marketing at Dictaphone. "It has emerged from the 'questionable' category and has begun to ramp up."

In addition to Dictaphone, companies battling for shares in this developing marketplace include Agfa Healthcare and MedQuist, a subsidiary of Philips, Provox Technologies, ScanSoft, Vianeta, eScription, Dolbey, and Crescendo.

Using the systems takes some time, as operators must learn to navigate the software and "train" the machines to recognize spoken words. The vendors claim that this can be accomplished in a matter of minutes and will bring extraordinary savings in time and money.

Dictaphone's PowerScribe reduced turnaround times by two-thirds and costs by $7000 per week in radiology departments at two hospitals run by Mercy Health System of Pennsylvania. The VR system saves $300,000 annually at Sarasota Memorial Healthcare System, an 828-bed regional medical center in Florida.

In the most common application of VR, known as "once and done," the radiologist dictates and then reviews the report, clicks an electronic signature, and sends the report. An alternative more in keeping with traditional workflow involves a "correctionist" who steps in for the radiologist to check the report for accuracy, comparing the dictation with the VR copy.

The technology is the same regardless of the model. VR vendors have built tracking devices into their editing software. This might be a bouncing ball on the computer screen that leaps from one typed word to the next in synch with the playback of the digital dictation. If a typed word doesn't match the spoken word, the correctionist or radiologist stops the playback, highlights the misprint, types the correct word, and goes on.

VR technology has matured through tools that assist in dictation and VR, as well as review of patient orders and demographics. MedQuist claims accurate speech recognition with its SpeechQ for Radiology as quickly as two minutes after a radiologist begins speaking. The longest time any company cites is 15 minutes.

"Smart fields" built into SpeechQ, released earlier this year, populate reports with RIS/PACS/HIS data and standardized text. Agfa's Impax TalkStation uses several dozen wild cards to fill in the blanks with RIS information.

"Any information that can be pushed to us from a RIS can be echoed back in these fields," said Dave Talton, Agfa solutions manager for reporting.

All VR products use just a few speech engines. Dragon Naturally Speaking, supplied by ScanSoft, is built into its own medical versions of the software, as well as those of PowerScribe and TalkStation. Philips' SpeechMagic uses SpeechQ for Radiology. (Until this year, MedQuist resold Agfa's TalkStation.) IBM's ViaVoice, which is the cornerstone of TalkNotes RAD from Provox, is licensed to ScanSoft for medical applications.

VR products are optimized for radiology self-editing, the most cost-efficient approach.

"This is the ultimate home run with the investment because you completely bypass transcription," Falatti said.

The systems also support the use of correctionists, because some institutions do not commit entirely to this model. They gradually transition to VR or maintain a mix of correctionists and self-editing, allowing radiologists to choose the alternative they prefer.

Hoping to cover all possible fronts, Dictaphone has developed an outsourcing model for institutions that prefer to use correctionists but not their own staff. The iChart service offers Internet access to the company's speech recognition and dictation software. Correctionist services, under contract to Dictaphone, edit VR reports.

Medical environments beyond radiology have welcomed VR, particularly pathology and cardiology, which, like radiology, have specialized vocabularies. It has also been applied as a low-end, voice-driven electronic medical record for digitally charting patient data across departments.

Dictaphone and other vendors are developing enterprise-wide products to meet these more expansive demands. The company's EXSpeech is designed to manage reporting workflow across medical specialties, hospital departments, and offsite clinics. Dictaphone has installed 200 such enterprise-wide systems, but that number represents only half its installed base in radiology.

VR first took root in radiology and it will likely flourish there first. The current technology has improved substantially since the systems were first introduced, but it has further to go. New features may decrease the time needed to create a report and may even improve the quality of the report by linking to other useful information. This would allow VR systems to automatically fill in historical and baseline information from previous radiology reports.

Dr. Lincoln R. Berland, vice chair for development at the University of Alabama at Birmingham radiology department, speculates that future VR systems may provide seamless connection to other information systems.

"They will close the loop with referring clinicians, making sure they receive the reports and telling you when they do. These are features that no human transcriptionist could match," he said.

Berland, along with 40 radiologists and 30 residents, uses VR to process a half million reports per year.