Raising Your Voice (Capability)
By Selena Chavis
Radiology Today
Vol. 23 No. 5 P. 22

Experts weigh in on how radiologists can get the most bang for their speech recognition buck.

Radiologists were among the first clinicians in health care to recognize the promise of speech recognition (SR) technology. As early adopters, many have watched the full evolution of SR in health care from a tool used to improve speech-to-text turnaround to broader applications made possible through advances in AI and cloud-based infrastructures.

Erin Palm, MD, FACS, vice president of clinical with Suki, notes that in the current health care climate, radiologists must fully capitalize on the power of SR to improve workflows. “When we think about the voice solutions that are available for radiologists, first of all, they’re 100% essential to their workflow because that time savings is so critical,” she emphasizes.

The realities of today’s radiology field are experienced by clinicians every day. Greater demand for imaging coupled with post-COVID clinician exhaustion and unprecedented workflow shortages mean that health care organizations must make the most of existing resources on the radiology front.

“Radiologists are in the front line of health care. Therefore, accurate and rapid turnaround of reports can ultimately improve the treatment and outcomes of patients with complex conditions requiring complex treatments,” says Jaymin Patel, MSc, MBBS, FRCR, chief operating officer of Hexarad, a UK-based teleradiology company founded by radiologists. “[SR] delivers the text accurately ‘verbatim,’ as errors can be avoided and, in some cases, lives saved. Users can deliver more reports in an accurate and timely manner, which, specifically in the current environment, helps with the backlog created over the past two years.”

Janusz Kikut, MD, vice chair for radiology informatics at the University of Vermont, believes that health care professionals who are putting off adoption of advanced SR will quickly find themselves behind the power curve. “There is a reason Microsoft spent all that money to buy Nuance,” he says. “Anyone who is not investing in voice recognition now, I think, is kidding themselves. And it’s not just radiology.”

SR technology, used the right way, has the potential to make the read-flow process more intelligent, meaningful, and accurate as well as reduce the stresses that radiologists find themselves under, says Shiraz Austin, cofounder and managing director of Scribetech Ltd. “With reporting done in a faster way, that helps tackle staffing shortages and improved workflows as report backlogs ease and reporting is provided instantly, leaving the radiologist to focus on treatment diagnosis and be part of the positive domino effect that faster and accurate detailed reporting provides,” he notes.

Kikut points out that SR not only streamlines documentation workflows but also enables clinicians to better understand and use software applications via a simple user interaction. “It used to be 50 commands that you really needed to do word processing. Now you have 1,000 other commands, and I think that prohibits many users from taking the full capability of the software and deploying it into their work. And, obviously, at the end it makes the product better and the user more efficient,” he explains. “[With SR], when you get the new software upgrade and all those features are hidden, you could just ask with voice or say, ‘change the font’ or ‘next page’ or whatever it is.”

Industry professionals agree that SR holds great promise for improving documentation and administrative workflows from many angles. Significant advances have paved the way for a heightened level of accuracy and, while they agree there is still a lot to be done in this area, radiologists will want to make the most of current SR advances to avoid mistakes and streamline workflows.

Best Practices
How can radiologists ensure they are laying the best foundation for value and getting the most return on an investment? Experts offer the following tips and considerations for deploying SR wisely.

Recognize the power of the cloud.
Palm says that the move to the cloud has led to dramatic accuracy improvements in SR technology. “For clinicians evaluating a voice solution for radiology today,” she says, “they’d want to make sure they’re using the cloud-based models rather than locally installed software because they get the benefit of this continuous improvement that’s delivered via the cloud-hosted software vs not getting access if it’s a model that’s installed.”

The cloud-based solution (Augnito) used by Hexarad has proven advantageous, according to Patel, due to the ability to minimize the need for health care IT teams to support the solution on site with additional hardware and or software. “Product application enhancements are now delivered centrally in the cloud as soon as they are ready for release, without the need for us to upgrade locally installed versions,” he explains. “This reduces dependency on IT resources teams as well as potential associated product downtime, delays, and costs.”

Deploy a specialty-specific framework.
In addition to cloud capability, Austin points to the importance of having a solution that understands radiology vocabulary, even when spoken with the strongest accents. This was an important consideration for Hexarad, an organization formed by radiologists to help radiologists.

In general, Austin points out that radiology requires less content but more specific words and phrases. Multiple subspeciality vocabularies are more widely used for a greater quantity of individual reports. Solutions that offer canned or standard text where diagnoses are normal and can be easily inserted with a single voice command are advantageous to improving workflow. “Other specialties often require more varied content, still detailed where the required report requires a more verbose response,” Austin says, citing patient notes and discharge summaries as examples.

Palm says the technical language that is familiar to radiologists is likely the easiest hurdle to overcome in a solution. When considering the workflow of radiologists, design matters. “If you think about somebody making rounds for patient care at the bedside—even interventional radiology is a different use case vs diagnostic radiology where they’re reading exams in a PACS, day in and day out,” she explains, adding that design will impact usability. “If you assume that mobility is not what they value, you’ll design interfaces for a larger screen design with more attention toward hardware, perhaps. These are just some of the factors that make diagnostic radiology a bit different.”

Consider the easiest path to implementation and optimization.
Industry professionals agree that addressing key factors on the front end can minimize headaches down the road. For example, sometimes adding new technology to existing infrastructure can limit the long-term potential of an SR strategy.

Palm says there are advantages to having a solution built on the latest technology from the beginning, especially when it comes to accents and accuracy. “Now, the models that are available can handle [significantly] more data to improve accuracy in all kinds of environments with different levels of background noise, for example. It’s able to incorporate different accents for English speakers within the same model because it can handle more volumes of data to train those models,” she says.

For Hexarad, reliance on a legacy SR system was a nonstarter, due to its long-term goals around efficiency. “It was cumbersome and required significant manual, offline manipulation by radiologists, which was costing them time. For health care providers and radiologists alike, this raised considerable issues with productivity, efficiency, morale, and patient safety,” Patel says, adding that because the legacy system was not an enterprisewide solution, changes had to be made on a local level, rather than just once on the central platform. “In addition, we had numerous complaints regarding the accuracy of the dictation from our radiologists,” he says. “We knew we needed to find a more reliable speech recognition solution; one that would be better suited to our enterprise and present a more user-friendly environment for our radiologists to dictate into and report through.”

Integration capabilities within an existing software ecosystem and potential extra costs are also important considerations, Palm emphasizes. Health care organizations should ask questions such as: Does it require custom hardware that’s different from what you’re already using?

Evaluate how SR incorporates context.
While advancements related to clinical context are still evolving, with new capabilities emerging on a regular basis, Palm suggests that health care organizations should identify and prioritize models that draw on the latest functionality to improve accuracy. Especially within the fast-paced workflows of radiologists, context is important to overall effectiveness.

“They are having to read so many studies under so much pressure, and they’re almost doing this in a vacuum and in isolation because they are pushed on throughput and incremental time savings,” Palm says. “Anything that can give them more context on the exam that they’re reading is tremendously important to their success and the relevance of their read.”

A Glimpse Into the Future
Looking ahead, experts believe the power of SR to impact workflows will continue to improve as capabilities related to context increase. Kikut says advances with AI will enable applications to more accurately differentiate between commands and identify what a radiologist actually wants in a report. For example, he cites the ability to differentiate “colon” as a punctuation mark and a clinical term.

Palm offers another real-life example that she believes could have been avoided by having better context. “We were seeing a patient who had a cardiac pacemaker, and one of the leads was causing a problem that resulted in local trauma,” she recalls. “The radiologist’s [note] said the pacemaker lead was traversing the uterus, which, one, is impossible that far away, and two, it was a male patient. So, you have these funny mistakes everywhere, but this is just one example of where the voice models being aware of that patient context could really improve the accuracy and relevance of the reads.”

Kikut points to advances with ambient capabilities that enable radiologists to dictate in a completely freeform, unstructured, and narrative manner while technologies create a fully structured report in the background. In addition, he believes that better incorporation of coding in SR solutions will continue to improve accuracy and context. Noting pancreatic masses as an example, Kikut says that variations of mass description require different codes. “All of the graphic features have different codes, and that is needed not just for the purpose of searchability and the findings in reports and the data, but also for the AI,” he says.

The bottom line is SR will continue to play an important role in the workflow of radiologists, and there is much to look forward to on the horizon. While the amount of screen time and repetition experienced by radiologists has made SR a natural fit with documentation for this specialty, industry professionals believe that many health care organizations have just touched the surface, in terms of how they can fully leverage SR in these workflows. The ability to retrieve and apply patient context is one area that experts agree will continue to deliver value ahead.

— Selena Chavis is a Florida-based freelance journalist whose writing appears regularly in various publications, covering everything from corporate and managerial topics to health care.