In July of 2003, the Sansum-Santa Barbara Medical Foundation Clinic in California replaced its five analog rooms with two full-field digital mammography (FFDM) units and one dedicated workstation, leveraging the existing general PACS for archiving and web access to digital mammograms and voice clip results for referring physicians and breast surgeons. The enterprise has been 100% filmless for digital mammography since go live with the FFDM units.

Nearly 2 years after initial implementation, access to the system by referring physicians has been well accepted, with a strong perception of increased productivity on their end. Mammography room productivity also is two and a half times greater.

Radiologist productivity, however, has been reduced by half, with annual reads of 18,000 mammograms taking twice the amount of time they did with film. Sansum’s radiology team is confident that a series of upcoming improvements to the system will help them achieve gains in radiologist productivity as well.

BACKGROUND AND BANDWIDTH

Sansum is Santa Barbara County’s largest group medical practice, with more than 145 physicians. The practice is affiliated with Cottage Hospital, a 366-bed facility, as well as two smaller hospitals. Together they offer a full range of imaging services, including CT, computed radiography, mammography, MRI, pediatric radiology, and ultrasonography. The combined professional radiology enterprise, including Cottage Hospital and the Sansum Santa Barbara Medical Foundation Clinic, is staffed by 12 radiologists who read approximately 220,000 examinations (CPT codes) annually. A residency training program at Cottage Hospital is staffed by the radiologists, with residents rotating out to the Sansum Clinic. Both organizations are entirely filmless and archived on two enterprise PACS from the same vendor, utilizing the same software and user interface, making it very efficient for radiologists and referring physicians to access either PACS database off of any LAN or web-connected device. The digital mammography system was the final modality added to the clinic enterprise, to result in a completely digital department. Cottage Hospital is completely digital as well, except for mammography, which is still film-based.

Using the dual monitor mammography workstation, Michael Trambert, MD, navigates a study.

According to Michael Trambert, MD, Sansum’s and Cottage Hospital’s lead radiologist for PACS reengineering, the FFDM implementation was “piggybacked” onto the existing PACS archive and web server. Examinations are interpreted on a dedicated digital mammography vendor workstation with two 5 MP displays and an additional PACS single-screen workstation for annotations and voice clips results logging. The annotation and voice clip results are available to the referring physicians on the PACS web server within 4 minutes of interpretation, along with all other radiology reports and examinations. Prior film-based mammograms are hung on an adjacent mammography viewer and are subsequently digitized onto the PACS.

Overall, the system gets a lot of use. Nine radiologists use the dedicated mammography workstation. There are four breast surgeons in the clinic, another eight breast surgeons outside the clinic, and about 500 physicians in the greater Santa Barbara area that all have user passwords to log onto the system from their office or home, and can navigate between the Sansum Clinic and Cottage Hospital PACS databases with the click of a mouse.

Trambert reports that most of the referring physicians review the mammographic images on good-quality flat-screen color monitors that they also use for general applications. Sansum Clinic breast surgeons have double monitors in their offices to enable viewing of left and right breasts on separate screens, as well as monitors at home. Access to the PACS also is provided in the Cottage Hospital operating rooms.

A key productivity component of digital mammography is bandwidth and adequate storage, Trambert says. Mammographic images are much larger than other modality images, roughly 45-50 MB per image uncompressed. The Food and Drug Administration currently allows lossless compression only for primary reading and archival of digital mammographic images. Lossless compression does result in an approximately 8 to 1 compression ratio, so that a 4-view screening examination results in a dataset of approximately 25 MB, which is the approximate size of an entire CT scan with lossy compression applied to the CT.

When Sansum first implemented digital mammography, it had a 100 MB network for the mammographic devices. Trambert says it took 10 to 35 minutes before the radiologist could even gain access to the study.

Six months ago, they took the entire network up to 1 GB. “It sped up significantly,” says Trambert. “Almost fourfold.”

The critical scenario wherein Trambert says the 1 GB network’s speed provides productivity gains for the technologist is in the add view or patient call-back workflow. In general, digital mammography tends to create fewer overall patient callbacks for additional views (calling patients back for magnification views or ultrasounds to work up questioned abnormalities) than film. However, when Sansum had its 100 MB network, the slow transmission time caused add view workflow to take anywhere from 10 to 40 minutes before the images were available for the radiologist to review and decide whether the add views were sufficient or whether further views were needed. The new 1 GB network has cut that delay considerably to between 2 and 10 minutes.

WORKLOAD BALANCING, VOICE CLIPS

Trambert warns that the dedicated digital mammography workstation cuts into productivity because it cannot be used for workload balancing. With all the modalities in the Sansum-Cottage Hospital networks, if one place is busier, the other place picks up some of the work. “We have four geographic sites that my group reads from,” he says. “They are all connected to the two different PACS servers, one at the clinic, one at Cottage Hospital. When I’m done reading one site, I can click on an icon in the upper corner of my screen and switch to the other server with the same user interface and application and start dictating cases at the other site.” With a dedicated system, he says, no one can share in the workload of reading the screening mammography because they have to be on the dedicated workstation to do it. Conversely, the radiologist on the FFDM workstation cannot use this device to read studies on the general PACS.

In an upcoming improvement scheduled for this summer, Sansum plans to eliminate the dedicated mammography workstation and will begin interpreting mammograms directly off the general PACS workstation, which will allow for workload balancing.

A key element that Trambert says increases referring physician productivity with the Sansum mammography PACS workstation is its built-in ability to do voice clips and provide annotations. If the study shows an abnormality, the radiologist uses the PACS workstation to create a montage image with annotation to help the referring physician more quickly see the abnormalities described in the voice clip.

The radiologist currently dictates the examination into a standard digital dictation system and simultaneously records the conclusion into the dictation system and into the PACS via a small microphone attached to the single screen PACS workstation. The montage and voice clip conclusion is permanently bound to the examination in the PACS, so when referring physicians bring up a digital mammographic study, they can click on a “play” icon and hear the radiologist’s conclusion while they look at the images. Later this year, both enterprises will convert to using an integrated dictation system from the PACS vendor that includes BI-RADS classification and direct integration with mammographic tracking software. This evolution will significantly improve radiologist productivity as well as office productivity in regard to mammographic tracking.

Voice clip results and annotated images are available on the general PACS Web server within 2 to 4 minutes after the radiologist has read the mammographic—or any other—examination.

“The physicians love it,” he says. “They really appreciate hearing the results. Being able to easily communicate our findings is a huge part of our value as radiologists.”

He notes that some PACS have incorporated the ability to type text notes in order to label images with results, but thinks that feature does not have the utility of a voice clip. “Typing is an extra step,” he says. “It slows me down. With voice clips, when I compare old and new studies, I can listen to the old conclusion while I’m looking at the new images and glance back at the old montage if what’s been said conjures up that need. The voice clip allows me to multitask.”

Trambert says he understands why an engineer designing a workstation might not think to incorporate voice clips. “You can’t search a voice clip, you can’t database it,” he says. “But as a productive radiologist, it’s a mission-critical tool.”

BETTER INTERFACE DESIGN

According to Trambert, the big opportunity in radiologist productivity gains will come from improving the user interface in digital mammography systems, integrating dictation, and BI-RADS integrated tracking. Current user interfaces, he says, are highly inefficient and contrary to the way radiologists work (see sidebar).

Sansum FFDM at a Glance

  • Mammograms per year: 18,000
  • Percent screening: 95
  • Radiologists: 9
  • FFDM Units: 2
  • Dedicated mammography workstations: 1
  • Average size of a mammography image: 45 MB
  • Average size of a mammographic image lossless compressed: 6 MB
  • Average size of a mammography study: 180-200 MB
  • Average size of a mammographic study lossless compressed: 25 MB
  • Network bandwidth: 1 GB
  • Perceived physician productivity gain: 100%
  • Mammography room productivity gain: 2.5 increase
  • Radiologists’ productivity loss: 2-fold

“Our interpretation time has almost doubled,” he says. “In a film-based environment with mammography viewers preloaded with examinations, old and new, the average interpretation time for screening mammography varied somewhere between 30 seconds and 2 minutes per study. With our current digital mammography system, we are between 2 and 4 minutes per study. Luckily, our PACS has made us so efficient at interpreting the other modalities, we have the staff to handle the extra work time.”

At the core of why it is more difficult to create efficient user interfaces for digital mammography compared with other modalities is that the images are so large. With a large breast, part of the tissue will lop off the edge of the screen and the radiologist has to pan around to see the edges.

Another core issue is the way most vendors present the examinations in digital scenarios that radiologists have to set up in advance, typically at least a nine-step process. When the new interface becomes available this summer, Trambert believes it will go a long way toward helping Sansum’s radiologists achieve the productivity gains that have so far eluded them with digital mammography.

“Radiologists in busy practices with an eye for increasing productivity and/or increasing accuracy should absolutely be involved with designing these interfaces,” he says. “It’s supposed to be a turbo Porsche, but right now, with the dedicated digital mammography workstation, it’s more like driving a bus. I think eventually, though, we’ll be able to interpret digital mammography studies almost twice as fast as what we can do now.”

After all is said and done, does Trambert think digital mammography is worth the expense and time? “It certainly is,” he says. “The images we’re getting are absolutely beautiful, and I’m fully confident we can realize the productivity gains we’re hoping for. Will we be twice as fast as an optimized film-based paradigm, which has been our experience with the other modalities on PACS? We’ll just have to wait and see.”

From 9 to 4 Steps

The key to radiologist acceptance of FFDM lies in the ability of vendors to make the interface less cumbersome, believes Michael Trambert, MD, lead radiologist for PACS reengineering at Cottage Hospital, Santa Barbara, Calif. Trambert says the following example is a typical workflow scenario in most FFDM systems:

1. New and Old, All Four Views. The new study appears on the right screen. The old study is on the left screen. Each study shows all four views in minified images, right and left breast views are back to back for comparison. The radiologist clicks to see the next view.

2. New and Old, Craniocaudal (CC) View. On the right monitor are the right and left CC, back-to-back and minified to fit both on the screen. The left monitor shows the same views of the old study. The radiologist clicks to see the next view.

3. New CC View. Now, the new examination takes up both screens, left breast on left monitor, right breast on right monitor, showing just the CC views back to back of the new study, typically not at full resolution. Radiologist clicks for next view.

4. New CC View in Full Resolution. Same new-examination CC views, right on right, left on left back to back, at full resolution, and in nearly three quarters of studies, part and sometimes much of the breast tissue lops off the screen, even at 5 megapixels. Radiologists use a panning tool to view the entire breast, looking for microcalcifications and asymmetries. When finished, radiologist clicks for next view.

5. New CC CAD Views. CC views of the new examination, minified with CAD marks superimposed. Radiologist clicks again.

6. New and Old Oblique Views. New examination on right monitor, back to back, left and right view in minified images. Old examination is on left monitor. Radiologist clicks.

7. New Oblique Views. Left breast on left monitor, right on right, minified again, back to back. Radiologist clicks.

8. New Oblique Views in Full Resolution. New examination. Right breast on right screen, left on left. The radiologist must pan around the images to see the full breast at full resolution in most cases. Radiologist clicks.

9. New Oblique CAD Views. Oblique views of the new examination, minified with CAD marks superimposed.

In this typical scenario, there are nine presentations of images. Other available scenarios are not appreciably more efficient, especially since diagnostic bilateral examinations are made up of 6 views rather than the standard 4-view screening study, resulting in a total of 12 steps as above.

Trambert has been working with a PACS vendor to develop a better user interface that will more closely mimic the way radiologists work with other digital modalities, and reduce the nine-step process to four. The interface will include:

Paging Format. “Everything is put into a paging format,” he says. “If you have multiple images in a series, you hold the mouse button down and you just roll the mouse and it rolls the images sequentially back and forth. It doesn’t require you to move your eyes because the images do the moving. That’s how mammograms should be read, too. Right breast on right monitor, left breast on left, and instead of clicking the mouse to go to the next view, just move it forward to go from the CC to oblique views.”

Improved Response Time. “The workstation’s response time should be instantaneous,” he notes. “Currently, there is about a 1-second wait when navigating from view to view. That doesn’t seem like much, but it’s very distracting; it breaks your train of thought. You don’t want any delay whatsoever.”

Hot Key. Trambert also wants a hot key that he can hold down, and when he starts moving the mouse, it would put the right breast image on the right monitor and the left on the left; not only would it take him on both monitors from CC to oblique view, it would interleave the comparison examination’s views in between the new examination views, facilitating comparison of imagery, all of it accessible just by rolling the mouse. Such a key would reduce the nine-view scenario to just four: New CC, Old CC, New Oblique, and Old Oblique. Then he wants to hit another hot key to see full resolution to look for microcalcifications.

Perceptual studies on presenting images support this interface because they show that people are much more sensitive and can more accurately detect subtle changes when the images are moved on the monitor in front of their eyes instead of the other way.

—T. Greenleaf

Tamara Greenleaf is a contributing writer for Decisions in Axis Imaging News.