Voice recognition software getting better
Voice recognition software getting better
Users say it will never answer all input needs
Lame typing skills are no problem for radiologists at LDS Hospital in Salt Lake City. With the facility’s voice recognition software, doctors can dictate their notes, watch their words take shape on the computer screen, and then edit them. It’s affordable, accurate, and fast. The turnaround from digital (computerized) X-ray image to report is usually under 20 minutes.
According to Roger Buxton, MS, RN, voice-driven radiology reports are just the beginning. Buxton, director of nursing information systems, sees LDS’ first application of the technology as the ideal way to watch it in action as it matures and spawns new models. In several years, he expects to introduce some descendent of this generation into nursing documentation. "Personally, I’m quite excited about the prospects of this happening," he says.
So far, staff at LDS like what they see, yet are not anxious to burden the technology with higher expectations than it can handle.
Among the early customers in radiology, satisfaction runs high. "One doctor recently told me that he gets a report back to the emergency room or one of the hospital’s outpatient clinics before the patient gets back there," reports Peter Haug, MD, LDS’s co-director of medical informatics. "As with any technology I’ve ever introduced, however, it has had mixed reception." Currently, 42% of LDS’ X-ray reports are completed via voice recognition. Haug expects to reach the 50% level soon. "There will be a subset of radiologists who never see the advantage in it," he says.
Other limitations of current voice recognition programs include:
- Confidentiality of patient information. While not such a concern in private office situations, it’s a particular problem with spoken data entry on nursing floors.
- Adaptability to stylized or structured data entry for extraction into quantitative reports such as outcome or utilization. Similarly, discrete data sets for the factual sections of the electronic medical record might present similar problems. "It will be difficult to get clinicians to comply with stylized input," Buxton observes. "We’re going to have to open the new frontier a little more first."
- Recognition of broad clinical vocabularies as used for internal medicine or nursing care. Currently, the software offers well-developed dictionaries for specialties such as radiology, cardiology, and emergency care.
- Accurate recognition and transcription when a user’s voice temporarily changes due to hoarseness or a bad cold.
Haug calls radiology report turnaround the "low-hanging fruit" among QI advances from speech recognition technology. "It will be harder in other areas where the clinicians have to move from one patient to another. The emergency room appeals, but we’ve not gone there yet. We are experimenting in the outpatient area."
Portable, voice-driven devices could be the next offshoot, Buxton says. Indeed, it would have to be something like that to beat the highly functional setup of bedside terminals on the nursing units. "We would have to make the new technology easier and faster. We’re not going to get them to lug a laptop in order to enter patient data as they move around the unit."
"Eventually, we’ll need a multimodal system, to combine voice recognition and keypunched commands," he predicts. The solution could be portable wireless devices. Perhaps pocket instruments akin to today’s hand-held computers. "We figure it will come on board in five to six years," he projects.
In the pilot stage, the vendor taught the most receptive radiologists how to dictate their reports to the computer. Now LDS employs professional trainers to teach the more reluctant group.
Training is in two stages. First, the user reads to the computer for 40 to 60 minutes, training the program to recognize his or her voice. "Then the system trains the radiologist!" Haug notes. Eventually, the radiologist learns to use proper diction and omit alien utterances like "um, ah, and oh well."
"After the system trains them, most of them like it. As we move, we ask ourselves how to make the incentives to use it equal to the disincentives," he adds. The disincentives involve about three weeks of dissatisfaction.
While several good software packages are on the market, Haug says, LDS uses the Dragon Engine. "The advantages of one system over the other are rapidly disappearing," he observes. "The performance of each will soon come down to the quality of a user’s computer and microphone or the bells and whistles within a software package."
For instance, at LDS, the Dragon Engine is embedded into a program called Radiology Workstation, which provides supportive data including the patient’s radiology history and other clinical data. This allows the physician quick reference to background information that can be useful in interpreting radiology findings.
With each new generation of software, training times and the consequent dissatisfaction period will shrink, he says. In the near future, he predicts that programs will recognize a user’s voice after 10 minutes. Accuracy of transcription will increase as well. "Every six months, there will be enough software improvement that we will want to upgrade our system for a while," he says.
Based on the LDS experience, Haug offers this cost information:
- Speech recognition package — approximately $500 per computer.
- Hardware requirement — $1,500-range computer.
- Development costs — installation time plus five to eight hours training for each radiologist. Training extends through the user-trains-machine and machine-trains-user periods.
(For a description of other information management programs at LDS, see QI/TQM, May 1998, p. 57.)
Need More Information?
For information on practical applications of voice recognition or dictation software, contact:
- Roger Buxton, Director, Nursing Information Systems, LDS Hospital, Salt Lake City. Telephone: (801) 408-1754. E-mail: ldrbuxto @ihc.com.
For information on voice recognition programs, contact:
- Dragon Systems, 320 Nevada St., Newton, MA 02160. Telephone: (617) 965-5200. Web site: www.dragonsystems.com.
- Lernout & Hauspie, 52 Third Ave., Burlington, MA 01803. Telephone: (781) 203-5000. Web site: www.lhsl.com.
Subscribe Now for Access
You have reached your article limit for the month. We hope you found our articles both enjoyable and insightful. For information on new subscriptions, product trials, alternative billing arrangements or group and site discounts please call 800-688-2421. We look forward to having you as a long-term member of the Relias Media community.