March 09, 2009

Talking It Through
By Kathy Hardy
Radiology Today
Vol. 10 No. 5 P. 10

Implementing speech recognition means change for the radiologists. Making it work requires managing that change well.

In Pete Townshend’s song, “Tommy, Can You Hear Me?” from the rock opera Tommy, the lead character becomes distracted by his image in a mirror. In the case of speech recognition technology, radiologists may rephrase that title to ask, “Tommy, can you comprehend me?”

“We recognize that speech recognition technology is not perfect, but neither was the old transcription method,” says Timothy McCowan, MD, radiology department chair at the University of Mississippi Medical Center. “It was like the PACS implementation. There was some pain but, if you embrace the new technology and go forward, it provides you with a tremendous advantage.

“In modern radiology,” he adds, “there’s really not an option to do things the way they were done before. Once we made the decision to go with speech recognition software there was no opting out.”

Going All In
The University of Mississippi Medical Center, a teaching facility located in Jackson, Miss., went live with Nuance’s RadWhere front-end speech recognition software in May 2008. The radiology department’s roughly 45 faculty members and residents made what McCowan called an “all-or-nothing” transition with a “drop-dead” implementation date for the switch from dictation and transcription to a speech-driven radiology environment. From the perspective of the medical center’s director of imaging services, W. Daryle Heath, BS, RT(R)(CT), this approach was the best way to establish smooth communication between users and this new technology.

“There was some push back from the staff,” Heath says. “But with the backing of the hospital administration and chairman, we moved forward. Within 45 days, the radiologists agreed that they didn’t know what they were missing without it.”

McCowan notes that without buy-in from the radiology staff, any facility will have difficulties with a complete transition. He says radiologists should be included in the decision-making process.

“At the end of the day, it comes down to the radiologists and what their pain threshold is,” McCowan says. “What other options do they have? Where is the relative pain? We’d like to get to the point where there is a core group of radiologists who want it and will prove that it can work.”

Heath and McCowan credit advance planning as the key to staff members getting up to speed quickly and utilizing the software’s features, such as creating macros for frequently used phrases and templates for regularly performed exams. They started the implementation with three months of advance planning and training. Radiologists received two to four hours of advance training worked into a two-week period. Trainers were also available on site the week the medical center went live with the software.
“That advance process and planning made the difference,” Heath says.

Dictation Surroundings
What users learned during training was realized as the software went into use, McCowan says. First, the medical center worked to create a dedicated area for exam reading that was out of heavy traffic areas, where noise from ringing phones or other disruptions could influence the speech transcription. Typically, a radiology reading room sees frequent activity, with staff members delivering charts and doctors consulting on cases. Any noise or discussion can become part of the dictation. Doctors need to speak well and be aware of their surroundings to optimize speech recognition effectiveness.

“You need some segregation of the reading stations,” McCowan says. “Some facilities can totally isolate their reading areas, but that’s not conducive for a training facility like ours, where we need to give instruction.”

Next, radiologists were reminded to use good diction as a way to reduce errors in their reports. Both McCowan and Heath say that speaking slowly and clearly are techniques that are stressed during training and should be used in actual practice. Users should speak with a consistent, deliberate speech pattern.

Languages can cause some problems as well; however, Heath notes that the software comes with language modules that can handle dialects from around the world.

“A common complaint is that the software doesn’t recognize certain speech or isn’t 100% accurate,” says Michael Mardini, vice president of imaging with Nuance. “It’s not perfect and the expectation should be set during the acquisition process. There’s an expectation that the software learns and that’s partially true. At the end of the day, very successful users learn to be deliberate in their speech patterns to optimize accuracy.”

Mardini recommends that radiologists read the exam, gather their thoughts for the report, and then speak slowly when dictating the report.

“Speak the way you would talk to your 5 year old,” Mardini says. “Be deliberate.”

Reducing Errors
Another mistake Winscribe support engineer Evan Armstrong sees radiologists make during dictation is that they speak in regressions, including discussions of various symptoms while dictating exam findings.

“This is not adaptive software,” Armstrong says. “Our brain can adapt to different language, but not speech recognition. Speech recognition is very literal. It will translate exactly what it hears.”

McCowan adds that there are typically more errors than users might think initially, and stresses that radiologists need to read their reports on the screen in real time or printed report form before signing off. He notes that when users speak the words, they often don’t recognize errors as they appear in their reports because in their subconscious, they see the word they said, not the word the software mistook for the correct word.

Building templates is another method that makes speech recognition easier to use in the long run. Some facilities will even form committees to build scenarios for templates in advance of the software implementation and then execute them once the system is up and working. However, it does require some time up front to implement, a stumbling block for some users.

“It takes some work on the user side initially,” Mardini says. “You can build a specific template for an exam and condition—a CT of the head for headaches in an adult, for example. As you continue to use the software, you can build and add more templates that you can automatically pull up when a case with that exam appears again. With our solutions, a properly planned set of templates can provide a standard, sensible, and structured look to a report that allows the radiologist to communicate findings freely in a flexible manner while providing a consistent, predictable look and feel for the ordering clinician.”

Customization vs. Standardization
While some users may customize features to meet their needs, creating macros for commonly used phrases, radiology administrators like Heath would prefer to see more standardization in usage across an organization. Creating templates and using them regularly results in a uniform presentation of more clear, concise reports, he says.

“As an administrator, I would like to see 100% standardization,” he says. “Some customization is OK, but I would like our reports to all look the same.”

From Armstrong’s customer support perspective, he sees reasonable customization as a way to help users adapt to speech recognition, either facilitywide or within individual client sites. He says it depends on the facility’s goals.

“For any IT administrator, it’s a matter of balancing the ability to support the new system while giving users the flexibility they like,” Armstrong says. “They don’t want every person doing something completely different.”

However, that can conflict with radiologists’ strong desire to work the way they are most comfortable. McCowan acknowledges that there is an art to reading images, and there will be instances where radiologists will customize their dictation techniques to fit their particular styles. But he agrees with Heath that standardization is important for maintaining consistency within the department.

“There’s a trend toward making everything more standardized, to have the same points in each report,” McCowan says. “Standardized reports that utilize more fixed text can also help reduce the number of errors in a report. Also, standardization is being stressed by various radiology societies as a practice all radiologists should follow.”

The University of Mississippi was using a dictation service when Heath joined the medical center staff, with a majority of the transcription done off site. The turnaround time in the month prior to implementing RadWhere was 24 hours for 80% of the work done. By the end of the first month postimplementation, the average completion time went to 18.6 hours, Heath says, with 96.9% of all reports signed in 24 hours. That percentage increased to 99.6% the following month.

With speech recognition, the medical center uses no transcriptionists. The entire process of dictating and editing exam reports is done verbally, with radiologists involved on the front end.

Radiologist Involvement
Each facility that transitions to speech recognition makes a decision as to how radiologists will fit into the process, and that can also impact how well the software initiation goes. For example, radiologists at the University of Mississippi all use the front-end approach and dictate exams and see their words typed as they speak. They then edit, authenticate, and sign reports before they are sent to physicians consulting on the case. The administration believes this method results in better quality reports because the images they are reporting on are in front of them. The facility also improved turnaround time.

Some radiologists contend that those gains come at the expense of the radiologists’ efficiency. With back-end speech recognition, a transcriptionist handles the editing and sends them back to the radiologist for approval. Back end is used when radiologists don’t edit their work.

“You can use both methods or one or the other,” Armstrong says. “We see mostly back-end users. We find that doctors don’t have time to dictate and edit exams.”

The ongoing debate surrounding speech recognition is whether the technology is good enough so that the change and/or additional work for the radiologists using it to report is offset by faster turnarounds and reduced transcription costs. Whether it is good enough often depends on who gets to make that call, because even with faster service and reduced costs, radiologists sometimes perceive they’re the ones shouldering the work while the benefit goes to the facility.

Historically, Armstrong says that radiologists in Europe were early adopters of front-end speech recognition, while U.S. radiologists were primarily back-end users. But he adds that he sees this trend shifting somewhat, particularly as younger radiologists join U.S. hospital staffs and imaging practices and as the technology continues to improve.

Mississippi’s Approach
The University of Mississippi Medical Center is a teaching facility with an influx of residents. This gives Heath a new audience of technology-friendly users who then become advocates for the system. Younger physicians tend to adapt to speech recognition software faster than the more established faculty on staff, he says. They use more technology in their daily lives on average and are often more willing to try new technology. On the other hand, with new residents arriving every July, Heath anticipates a short period of adjustment that might negatively impact report turnaround times, but estimates that within 45 to 50 days, those times should return to normal.

Heath developed his best practices for implementing and using speech recognition software with his previous employer, where the decision was made to ease the radiology staff into the new software.

“If you’re implementing new technology, it’s to decrease costs and increase output,” Heath says. “By not going 100%, you shortchange yourself. Some physicians will adopt it and some will not. Or they will utilize speech recognition but use it as dictation and send the report to someone else for corrections.”

Regardless of how it’s used, speech recognition technology is finding its place in the electronic communication systems of hospitals and imaging facilities with various degrees of success.

“It will be exciting when the day comes that you can just talk into a machine and produce a report,” Armstrong says.

— Kathy Hardy is a freelance writer based in Phoenixville, Pa., and a frequent contributor to Radiology Today.