CHICAGO – At RSNA 2011 earlier this week, David L. Weiss, MD, of Carilion Roanoke Memorial Hospital in Roanoke, Va., asked for a show of hands.
How many people in the audience were radiologists? Almost all of them were.
And how many of those people use speech recognition technology in the course of their work? More than half of them raised their hands.
Finally, how many of those people were completely happy with the technology? Just two or three of them were.
Speech recognition is a technology with a lot of potential – but many pitfalls. Sometimes its voice-to-text translations are nowhere near as accurate as one would like – making more work for already-busy clinicians as they go back and correct mistakes.
[See also: Speech recognition leads to imaging report errors, study says.]
There's "no correct way to use it," said Weiss. That said, some uses are more efficient than others – and some can be truly innovative.
Weiss, with more than a decade of experience using the technology, offered some tips for optimizing speech recognition use.
- Dictation style is crucial. Weiss suggested speaking in continuous and complete phrases, since the technology relies on the context of individual speech phonemes to glean meaning. Day-to-day consistency is crucial too, as most systems will evolve and adapt to your speech habits. And deeper voices are better, said Weiss. If you're having trouble with your dictation program, he suggested channeling your inner Sam Elliott.
- Corrections are key. Using functionalities like highlight-and-repeat, train/correct and add/change/delete is another way to mold one's speech program to one's habits and speaking style, said Weiss. He mentioned how his software kept mistaking the phrase "one of these" for the term "wannabes." So he just deleted the term from its vocabulary. He didn't have much cause to use the word "wannabe" in his clinical work after all – except, he joked, to say that "cardiologists are wannabe radiologists.
- Microphone timing and sensitivity make a difference. Weiss reminded the audience that there's a split-second delay before the mic kicks on. He also pointed out that its sensitivity is variable, depending on the user's workstation, background noise, time of day, etc.
- How you navigate matters. Weiss put it simply. Use: voice commands and user interface devices. Avoid: icon clicks, drop-down menus and keyboards. Avoiding distractions from image diagnosis is crucial, he said. To keep his eyes – and mind – on the image, rather than the speech rec software, he tried out several different UI peripherals, including a handheld mics, a multi-button mouse, a ShuttlePRO controller, and more. He suggest using trial and error to find what works for you.
- Smart use of macros and templates can help. They're critical to efficiency, said Weiss, and very easy to modify and to remember. He demonstrated how he had easily customized his program to insert time and date stamps, and to adapt to a modality/body part/diagnosis format ("chest PA and lateral normal," "ultrasound fetal normal," etc.)
[See also: Speech recognition market poised for growth.]
Early in his talk, Weiss had teased the audience by promising to depart from custom and reveal who the specific vendor he considered best in the speech recognition field.
And so he did. His answer? "The other guy."
The grass will always be greener on the other side of the fence, he said. You'll always have a radiology colleague who boasts about a 150 percent accuracy rate. The answer, instead, is to use the technology that's right for your practice.
Don't decide based on speech engines, said Weiss – they're mostly the same and will evolve at roughly similar rates in coming years. The key, he said, is to do your homework, pick the vendor that will best jibe with your workflow, and, of course, make sure the technology is interoperable with your RIS/PACS.
