NYMISS: PACS needs to get smarter
September 14, 2012
by Brendon Nafziger
, DOTmed News Associate Editor
Radiologist assistants at the Mayo Clinic in the mid-1990s helped make the doctor's life a lot easier: they would hang current and comparison scans the way the radiologist liked, take down transcriptions in real-time and question the doctor if something seemed amiss, track down referring clinicians if there were urgent or unexpected findings and make sure they were followed up on.
And swapping these helpful "Mrs. McGillicuddy's" for picture archiving and communications software and voice recognition programs in the modern digital radiology suite has in some ways been a poor tradeoff, says PACS pioneer Dr. Elliot Siegel.
"It seems we've taken a step back," Siegel, chief of imaging services at the Veterans Affairs Maryland Healthcare System in Baltimore, and a professor at the University of Maryland, said during a talk at the NYMIIS at a Times Square hotel on Monday.
Siegel turned the VA Baltimore into the first filmless U.S. hospital in the early 1990s, and now as the health system looks to replace its original PACS with a next-generation sequel — the new vendor should be announced in a matter of weeks — Siegel thinks current PACS need to get smarter.
Siegel said one of the core problems with PACS — and a problem not shared by the radiologist assistants he worked with during his time at Mayo Clinic in Rochester, Minn. — is their inability to learn.
For instance, the way it hangs current and comparison studies is "really, really dumb," he said, as rather than learning from his preferences it requires him to adjust the layout every time.
Also, he experiences this when he starts his day, as his PACS won't let him sign on unless he uses all capital letters for his sign-on name -- something he often forgets to do.
"It keeps asking me questions like, 'Are you sure?' although I've been saying yes to that (issue) for the last 19 years," he said.
Closing the "communication loop"
Possibly one of the biggest issues with PACS is it could be doing much more to help radiologists document that referring clinicians follow up on suspicious findings, Siegel said.
Currently, for as many as one-fifth of cases there's no follow-up even if the doctor acknowledges receipt of the radiology report that contains the follow-up recommendation, he said. While the doctors' decision to not follow up could be medically justified, the radiologist should be able to quickly check if his recommendations slip through the cracks.
And closing the "communication loop" could involve automating a process so radiologists could see if their recommendations — such as a CT scan after a suspicious chest X-ray finding — were actually acted upon.
"I have no idea if people are following those recommendations. I want that in the next-gen systems," he said.
"Brain dead" speech recognition
Siegel would also like voice recognition programs, which transcribe the radiologist's reports, to take a smarter approach in order to avoid some of the hilarious misunderstandings he and his colleagues have had to put up with.
For instance, he said software has rendered the finding "fibroids of the uterus" as "fireballs of the Eucharist," the measurement "4.8" as "foreplay," and the phrase "if clinically indicated" as "death clinically indicated."
"Speech recognition systems are brain dead," he said.
In the same way a Mrs. McGillicuddy, a skilled and experienced human transcriptionist, might question the use of an unusual word, Siegel would like the system to call attention to one-of-a-kind words ("Eucharist"), or at least leave them blank, so the physician could correct the system's errors in review.
Siegel also hopes radiologists are able to, eventually, move past the dominant PACS interface for the past two decades: the computer mouse.
"I want some more degrees of freedom to do more complex navigation through datasets," he said.
One immediate alternative is a roller mouse. Siegel said when his department turned to using a roller mouse, they achieved a 40 percent reduction in CT reading time, because they could drag images continuously with the roller.
Touch screens offer another way to achieve some workflow shortcuts. The drawback is that radiologists probably don't want to be getting fingerprints on a monitor they have to make clinical diagnoses off of. That's why another soon-to-be-released product, the Leap Motion, could be interesting, Siegel said. The $70 gadget, about the size of the iPod, helps transform screens into "no-touch" touch screens by capturing fine hand movements. An early version of the device could ship as early as February, according to the company's website.
A possibly more sci-fi future interface brought up by Siegel is a $299 EEG headset produced by Emotiv, called the Epoc, that lets users control a computer by reading changes in brain signals. Reviews online suggest it has a few kinks to work out, at the very least, and Siegel also suggested it looked like it needed some work. But the idea holds promise.
"I don't think we're ready for this in 2012, but the idea of being able to refine this seems interesting for the future," he said.
Another problem with PACS is that no software developer provides the "best of breed" product in every category, but it's hard to get the systems to work with programs designed by other vendors.
"Our PACS vendor told us we couldn't put speech recognition in the system as it would void the warranty," Siegel said.
As a result, he said there's a lot of shuffling from one work station to another. But he would like a PACS that acted like an imaging apps store and could host modules or programs from a variety of vendors: the best virtual colonography program, the best ultrasound viewer, the best CAD for mammography, the best cardiac CT.
"Having that type of app store is something we'll see more of in the future," he said.
Watson in medicine
Siegel also hopes this year is the year artificial intelligence makes an impact in health care. And using AI for radiology is something he's working on with IBM, consulting with them on ways to get its Jeopardy-playing Watson computer to translate over to medicine.
The so-called DeepQA technology can process 500 gigabytes a second -- the equivalent of reading nearly one million books per second, he said.
With artificial brain power like that Siegel envisions a sort of virtual radiology resident or fellow, able to do some of the preliminary work for the radiologist.
"Wouldn't it be amazing (from a) clinical perspective if we had that capacity?"