Speech recognition technology continues to improve

Gerald Roth, MD, Henry Mayo Newhall Hospital & Tower Imaging, Valencia, Calif, working with one of Nuance’s speech recognition systems.

Digital imaging and radiology information systems have created an environment in which images from any number of patient exams are available immediately. Thanks to the shrinking number of available transcriptionists, the final report is often far behind—in busy hospitals, it can take as long as several days—putting radiologists in a pinch.

“Primarily, we were looking for a solution to really decrease our turnaround time for reporting,” said Douglas Fulk, MD, president of Cedar Court Imaging, Carbondale, Ill, who echoes the sentiments of radiology departments and stand-alone imaging centers that are turning to speech recognition (SR) technology to expedite report delivery. “We wanted to get our reports out in a timely manner, preferably the same day, and we thought voice recognition would provide us with that capability, 365 days a year.”

The Need for Speed

The good news for those eager to employ speech recognition—also referred to as voice recognition—technology is that it delivers.

“Voice recognition software has gotten significantly better over the years. You actually now have the ability, in some vendors, to have multiple voice models that are more targeted to the type of work you are doing,” said Sean Moynihan, director of Radiology Information Systems for the University of Virginia Health System. For example, the system will take into account that interventional-type procedures are more surgical than standard radiology work. “Today, clinical notes that radiologists dictate are just like writing a letter. They have to train the system and work with it, but there is a lot of functionality for users.

For many clinicians, their first experiences with SR was far less enjoyable. The technology initially suffered from overpromotion, and in many cases it simply did not live up to its promise.

Recent years, however, have brought advancements not only in the engines that run the speech conversion, but also in the computers that are now able to run these complicated algorithms at the speeds necessary for them to be effective.

While “once bitten, twice shy” may be a wise adage, SR seems to be a case where a second chance is warranted. Those who have worked in the past with systems that were either painfully inaccurate or completely hopeless just might want to try again.

“What I hear over and over is so many experiences with voice recognition where radiologists are generally displeased with it. The problem is there are a lot of bad, inefficient systems out there, so it’s critical that [facilities] bring a quality system in to show them how effective it can be,” said Brandon Selle, practice administrator for Northeast Missouri Imaging Associates, Hannibal, Mo. “A lot of radiologists have probably used bad PACS systems, or they’ve had to work with bad equipment, and it’s no different with voice recognition—there are bad systems out there, and there are good ones.”

Today’s software goes beyond algorithms that simply create text from audio input. Most solutions include a variety of shortcuts designed to help save time. Templates, for instance, allow radiologists to set up generic “normal” reports that require very little editing before signing off on them. Another common tool, macros make it possible for the clinician to utter a single word and have full sentences or paragraphs be inserted into the report.

“We are an academic practice and a big hospital, so there aren’t too many normal tests, but it’s nice to have the template, because it can help you organize your thoughts,” said Matthew J. Bassignani, MD, associate professor of radiology, University of Virginia Health System, and section chief of genitourinary imaging. “Once in awhile you actually do run across a normal, so the report is already dictated—you just sign it and move onto the next one.”

Exactly how much functionality is included with a system—and how user friendly it is—varies widely by vendor.

“A number of my colleagues would like to customize the way some of the macros can be created,” said Joseph Stock, MD, chairman of radiology at Crozer-Chester Medical Center, Upland, Pa. Describing the manual process involved with setting up new macros, he added, “As one of my colleagues noted, it would be nice to be able to alter your macros with voice recognition.”

Once mastered, the tricks and tools included in the SR packages can produce some very tangible improvements to productivity.

“When we installed speech recognition, we were doing about 130,000 cases a year with seven radiologists. Last year, we did 164,000 with five radiologists,” said Howard Stewart, the RIS/PACS administrator at Southern Ohio Medical Center. “So, obviously, they are more productive now, but that didn’t happen overnight.”

Making the Move

One thing is clear when speaking with those who have implemented SR: the choice to bring it into a practice or imaging department must include the radiologists. Failure to loop them into the decision-making process could mean failure for the project as a whole.

“There are lots of reasons why people would be resistant—they believe it’s going to add time to their day, and it does add a little, but not that much. So it’s important for radiologists to be part of the process to move to voice recognition,” said Bassignani, who is also medical director of Radiology Information Systems, as well as medical director for UVA Imaging Centers. “We hear plenty of stories of a hospital that installs voice recognition and the radiologists reject it, so they end up spending a half million dollars for nothing. I think that is poor utilization and poor planning, and I think it falls squarely on the shoulders of those people who excluded the radiologists from the decision.”

Part of that decision-making process is determining exactly how the software will be rolled into the imager’s workflow. There are essentially two choices. In the first, radiologists can dictate into the SR technology, then forward that audio file—along with the preliminary text document—to a transcriptionist who will listen to the dictation while following along on the document, making corrections as necessary.

The other option requires radiologists to dictate, review, edit, and sign the report in a single sitting. As previously mentioned, many doctors feel this is extra work that will increase the amount of time they spend reading cases.

“There is a trade-off, but it’s probably a wash. For me to dictate the report, review it, and sign off on it, those steps are slower than just dictating and going to the next case,” Fulk said. “But if I just use conventional transcription, there is going to be a time where that report will come back to me. I’m going to have to proof it, which takes time, and if there are errors and I don’t remember the patient, then I have to go back and review the images again. Whereas if I proof my report with the images in front of me, there is no time lost: once I’m done with it, it’s gone forever.”

Having a report done in one sitting can be appealing to radiologists, but many facilities don’t realize true enlistment with SR until the positive feedback starts rolling in from customers.

“For our radiologists, one of the positive reinforcements was the referring physicians coming back and saying things like, ?Your report was waiting for me even before I saw the patient in the office today’ and our doctors really responded to that,” said David S. Mendelson, MD, director of Radiology Information Systems at Mount Sinai Medical Center, New York. “People tend to view radiologists as an isolated specialty, but there is enough interaction with groups of physicians that were very complementary.”

Medical professionals in other departments often can see benefits beyond simply receiving their reports in a timely fashion. At the UVA Imaging Centers, one oncologist informed Bassignani that he was able to schedule more patients in a day, thanks to the immediate availability of the reports.

Before the facility implemented SR, the specialist would schedule 15 additional minutes between patient visits to allow time for the collection and organization of patient charts, specifically, the final report from radiology. Now that the report is invariably in the RIS with the images, he no longer has to include that buffer.

“He is more efficient now because we made our practice more efficient, and that’s a big boon for our entire enterprise,” said Bassignani. “He is an oncologist who is seeing more patients per day now because I am doing my reports with speech recognition.”

Justin Tholany, MD, Memorial Hermann Healthcare System in Houston, using one of Nuance’s speech recognition systems.

No Uniform Agreement

Even with ample advanced warning and active involvement, there is no guarantee that radiologists are going to welcome the myriad changes a SR system brings with it. Exactly who will welcome it, though, is often unpredictable.

“Will I tell you, everybody loves doing it this way? No. I will tell you that the residents eagerly embraced it; they just universally accepted it,” Mendelson said. As for the attending staff, he estimates that one third would say they love it, one third could take it or leave it, and the remaining third most likely prefer the old way. “There is some slight correlation with the age of the physician and how willingly they adopt it, but some of the doctors who most actively embraced it are older—so that is not a uniform observation either. You find that those who are more entrenched in traditional workflow are just having change management issues.”

In some cases, the type of work and when it is taking place have an impact on how eager radiologists are to employ SR.

“Some radiologists have resisted using it because it slows them down, and part of that is based on how busy we are. If we are reading CTs on-call at 8 pm, it is difficult to make the corrections that are necessary for the final report, so in that situation some people send it to transcription,” Stock said. Even then, approximately 85% of the center’s reports are completed with SR. “Our administration would like us to go to 100% because of the cost savings, but currently that would be difficult for us to accomplish.”

One of the biggest changes in workflow for organizations opting for the “dictate and sign” approach is that radiologists are required to do work that was previously handled by transcriptionists. Radiologists are often not hesitant to address this issue.

“There is the perception and, to some extent, the reality that the health system is saving hundreds of thousands of dollars—but it’s causing more work for radiologists,” Stock said. “But it is primarily a culture change, and we are urging people to use the system more by providing some incentives to make that a higher percentage.”

At Southern Ohio Medical Center, when doctors get frustrated with the SR system, Stewart encourages them with a somewhat cunning strategy.

“As the frustration increases, the recognition decreases, so on bad days we just tell them to get up and take a break. When they are gone, we copy a backup of their speech profile onto the system,” he says. While this does help “clean up” the accuracy to some extent, the biggest benefit is the 30 minutes it takes to complete. “They have time to get a cup of coffee and calm down a little bit, so when they come back and sit down to dictate, the recognition is much better.”

A Program, with Benefits

While many facilities wrestle with the best ways to encourage use of SR systems, Northeast Missouri Imaging Associates took a totally different approach to the situation. The group’s doctors made the decision to purchase the SR technology on their own, tacking on a service fee to every case read for their clients.

“Our physicians bought into speech recognition, because they literally bought into it. It was our money invested into the system; they are the ones who exclusively chose this,” Selle said. Before implementation, the hospitals and outpatient clinic Northeast Missouri Imaging reads for were handling all report-generation duties and were experiencing all the familiar issues. “We were able to offer the final reports to them at a reduced expense, compared to either in-house or outsourced transcription, so they were willing to embrace it,” he said.

The result has been a win-win situation for both the imaging center and its referring physicians. A turnaround time of about 2 days was clipped to an average of about 20 minutes.

“We were looking to break even on our investment and come out ahead on our service,” Selle said. “Not only do we return reports more quickly, but they are actually paying us less than they were paying before, so they were able to cut transcription costs.”

Another benefit for referring physicians is an increase in accuracy. Though the reports will sometimes include typos, often a result of editing that was not quite meticulous enough, many facilities have seen a decrease in more critical mistakes.

“We had some experiences with transcriptionists who weren’t familiar with infrequently used terms, and who would enter incorrect words—and those become clinically significant errors,” Stewart said. “Speech recognition, however, can be trained to learn how a doctor says a particular word. Even if he uses that term only once every 3 or 4 months, the program types it correctly—so our risk management group is also very pleased.”

Where It Counts

While the SR technology on the market today is far from perfect, the manufacturers responsible for its performance regularly release upgrades designed to make it better and more user-friendly. In the interim, clinicians who are committed to making the most of the programs do see rewards—not only monetarily, but also through improved patient care.

“All our referring physicians really want is the report as fast as they can get it, and speech recognition helps us in providing a much better service in a more timely manner,” Fulk said. “When a patient calls, the doctor no longer has to say they don’t have the report yet, so it makes everybody look good. Of course, it makes me look good by getting it to them quickly, but it also makes the referring physician look good, because when the patient does call, they can give the answer.”

Dana Hinesly is a contributing writer for Medical Imaging. For more information, contact .