Mammographers ask vendors for consistency in digital display

Digital mammography cleared a substantial hurdle earlier this year, when researchers established its clinical advantages in women under 50 and in those with dense breasts. Now, with demand for digital already on the rise, the technology faces another obstacle: image integration and workstation interoperability.

Making the transition from analog to digital can be tough enough by itself. But, in addition, digital mammography image display can vary widely, depending on the devices used for image capture and soft-copy review. Incompatibility between vendor systems and third-party workstations means that digital mammography images cannot be viewed side by side with MR images or ultrasound scans on a PACS. And sites that have more than one vendor's digital mammography system must use two different workstations in order to read exams.

If not for a five-hour forum held in June during the Society for Computer Applications in Radiology meeting, these problems might have festered unaddressed. But since that event, with the prodding of a group of pioneering users at the Elizabeth Wende Breast Clinic in Rochester, NY, a new task force has taken shape through the Integrating the Healthcare Enterprise initiative to tackle digital mammography integration.

The ultimate goal is the ability to read and interpret digital mammography images on any available PACS workstation, said J. Anthony Seibert, Ph.D., a professor of radiology at the University of California, Davis and immediate past chair of SCAR.

"Digital mammography technology needs to catch up to its clinical promise," Seibert said.

Each of the commercially available digital mammography systems is different in many ways, from detector aperture and pitch to x-ray conversion and image matrix size. Image postprocessing, which typically relies on proprietary algorithms intended to give a vendor an edge in image display, also varies. Even so-called raw images are not what they seem, Seibert said.

"One might think that raw images are the direct result of the way the x-rays interact in the breast and how that information is presented to the detector," he said. "In reality, many preprocessing steps occur to create a 'raw' data image."

At Elizabeth Wende, which uses digital mammography units from three vendors, radiologists embarked on a connectivity project. The intent was to show that images from any one of the vendors could be moved, intact, with full integrity and be displayed on a soft-copy review workstation.

Two years later, despite close collaboration with manufacturers, the goal has been achieved with only one of three vendors, said Dr. Margarita Zuley, a radiologist at EWBC.

As a result, EWBC radiologists must contend with a different interface for each vendor's equipment. Technologists and radiologists have a limited ability to communicate with each other electronically using current systems. Acquisition units cannot display annotations placed on the images at the soft-copy review workstation in order for technologists to perform extra views. Nor do hanging protocols for routine extra views exist on the acquisition units.

Interpretation can take up to two times as long as that for screen-film because not all of the data are available in one hanging, Zuley said.

"In order to see all four views at once, the images are displayed to fit the window," she said. "But to look for subtle lesions, full resolution is necessary. That means radiologists must review multiple display renditions of the same mammogram to arrive at a diagnosis. This breaks concentration, because their search patterns, developed through reading analog films, are disrupted."

There may be only a handful of sites across the country that use equipment from multiple vendors. So why rush to solve a problem that may affect only a few users?

In actuality, any site that reviews mammograms sent from another facility will be affected if the referring site does not use the same vendor's digital mammography device. Lack of integration between digital mammography units and third-party soft-copy workstations locks facilities into buying expensive proprietary workstations for diagnostic display.

Digital mammography is already a pricey proposition, with devices ranging from $300,000 to $500,000, compared with about $100,000 for a screen-film system.

"If you tack on a workstation that you have to buy because it is the only one compatible, that's another $75,000 to $100,000," Zuley said. "And then you have to deal with your PACS, because that digital mammography soft-copy review station might not talk to your PACS. It might not be compatible with your archive."

In addition to the financial costs involved, a price is paid in workflow. For example, images acquired from one vendor's equipment cannot be directly compared with those acquired by another's, she said.

"For workflow reasons, for budgetary reasons, for patient care reasons, you should be able to take an image and display it on any workstation," she said.

Users hope that the ability to read digital mammography images on universal workstations will put an end to "knobology," the science of learning how to use all the knobs on new digital mammography workstations. The term, coined by Dr. Dianne Georgian-Smith, a radiologist at Massachusetts General Hospital, aptly describes the conundrum mammographers face in sitting down to interpret digital images.

VENDOR CHALLENGES

Manufacturers and PACS vendors have risen to the challenge of providing a seamless setting for digital mammography review, Zuley said. One obstacle has been the tortuous regulatory environment for mammography manufacturers, which is unlike that faced by other modalities. The premarket approval process for digital mammography is costly and time-consuming.

"The vendors are not creating this situation by themselves," Zuley said. "They are very constrained by government. Every time vendors want to make a change, it can take up to a year and half, and tens of thousands of dollars. We need standards in place so that we have quality, and we need to allow our vendor colleagues to help us."

Vendors have been working on the problem for some time. Hologic has decoupled image processing from its soft-copy workstation and shifted that functionality to its acquisition device. As a result, the company's digital mammography systems can send full presentation images to any DICOM-calibrated workstation for viewing, according to the company.

GE Healthcare is doing its part as well. Users say one drawback of GE's digital mammography unit is its small field-of-view, which requires multiple views in women with larger breasts. The company was expected to introduce a new digital detector at the RSNA meeting with an active area of 24 x 30.7 cm to alleviate the need for multiple views.

The company is also addressing workstation issues, said Marcia Hill, clinical research and development manager for GE.

"We are implementing all of the necessary tools on our workstations to make sure we can display any vendor's images," she said.

With the September results of the national Digital Mammographic Imaging Screening Trial favoring digital over screen-film, such efforts have become a business necessity. Only about 8% of installed mammography units are digital.

"I haven't heard anything from vendors but a willingness to work with us. But it's not entirely altruistic," Zuley said. "There's a big market share to be had once these issues get figured out. Vendors know how much of a headache this is going to be when everyone is reading digital mammograms and women move across the country with their prior images on a CD."

GOALS AND PRIORITIES

At the first meeting of the IHE digital mammography task force in July, members defined top priorities and long-term goals (see accompanying chart, "Integration issues in digital mammography.")

The 20-member task force includes radiologists, physicists, and representatives from PACS and digital mammography vendors. It will first tackle the relatively easy to fix inconsistencies in image presentation, such as size, Zuley said. Currently, image sizes vary because the pixel matrix of each vendor's detector is different, ranging from 100-micron pixels (GE) to 50-micron pixels (Fischer).

"We are trying to get all vendors to display true size," she said.

The task force will also address ways to achieve consistency in image label presentation and to ensure that the blackness of image backgrounds remains the same across all images.

"This is all basic stuff that has already been defined in the DICOM standards, and it should be easy for the vendors to implement," Zuley said. "It's a starting point."

The task force will work on implementing IHE profiles that are already in place for general radiology and applying them to digital mammography. These initial tasks will be tested at a January 2006 IHE connectathon, Zuley said. A follow-up event next spring will involve a hands-on presentation in which radiologists can evaluate the progress made.

"Down the road, we will be writing profiles in IHE that are specific to mammography," she said. "The biggest clinical issue is getting the images to look the same when you take them from the vendor's own workstation and put them on another's-so that they can be compared side by side."

It's not clear how vendors will tackle the underlying issue of image inconsistencies: the processing of "raw" data. In DICOM parlance, digital images are considered either "for processing" or "for presentation." Images considered for presentation are already rendered and processed, with linear enhancement, contrast enhancement, and edge enhancement applied. The resulting image is one that any PACS-calibrated monitor can display in the manner in which it was meant to be displayed. Images in the for-processing category are those not subjected to any secondary postprocessing.

"The concept we are working toward in digital mammography is that all for processing images be the same from vendor to vendor," Seibert said. "But that is complicated by the fact that not all of these vendor devices have a linear response to x-ray exposure."

That's important because in order to create images that will display similarly, the linear response to incident exposure at the detector must be measured. In addition, variations in gain corrections, also known as flat-field corrections, must be implemented. The only way to get from here to there, Seibert said, is to remap all of the so-called raw information by measuring each manufacturer's response to incident exposure.

The map would provide a model against which to evaluate true raw images that are directly related to incident exposure and bit depth.

Whether the IHE task force will adopt that model as its means to achieve true image consistency remains to be seen. Such an approach, or one similar, is necessary, Seibert said.

"Ultimately, the IHE will have to address the issue of multimodality image viewing workstations and the ability to display and review mammograms in the appropriate hanging protocols simultaneously with other images. We are very far away from that right now," he said.

STICKY WICKETS

The sticky wickets in all of this are the proprietary algorithms that vendors apply to make their images look better than their competitors'. These range from edge enhancement algorithms that create sharper lines to tissue equalization that smoothes tissue. To achieve true image consistency, vendors may have to share those algorithms.

"The effort to achieve image consistency is going to neuter vendors a bit," Seibert said. "That's why it's going to be very difficult to get where we want to go."

Pressure to reveal behind-the-scenes processing may also come from users, who have expressed concern about the effect it has on image data. Radiologists are not provided with detailed information about either the algorithms used or their effects on the image (Radiology 2005;234:353-362).

It's also unclear which image processing algorithm is best. In a 2000 study that evaluated eight image processing algorithms, Dr. Etta Pisano, chief of breast imaging at the University of North Carolina, Chapel Hill, found little consensus (Radiology 2000;216:820-830). Radiologists preferred different versions of processed images depending on the task (screening versus diagnostic mammogram), lesion, and device type.

"We really don't know at this point what processing algorithms contribute to detecting cancer," Zuley said. "Will one processing method allow me to see calcium better, or is one better in dense breasts? We don't yet know. We are first trying to eliminate all the annoying variables like size differences so that we can create comparable images. Then we can evaluate the processing. We hope the vendors will help us figure this out."

Understanding such image processing tools will become more important as integration efforts proceed, said Elizabeth Krupinski, Ph.D., an assistant professor of radiology at the University of Arizona.

"It's not just a matter of give me a workstation that will display everything. It's give me a workstation that will display everything in the optimal manner," she said at the SCAR forum.

The newly formed IHE digital mammography task force will continue to work on long-term image integration goals during quarterly meetings. The process is expected to take years to complete. In the interim, digital mammography purchasers should request that IHE profiles be implemented on devices ordered or at least require that such compliance be part of the request for proposal.

"Because of DMIST, there are going to be a whole lot of purchase orders placed for digital mammography," Zuley said. "Facilities should make sure to request this in order to protect themselves, so that when the IHE profiles are met, they can get upgraded."

Ms. Dakins is feature editor of Diagnostic Imaging.

Integration issues in digital mammography: short- and long-term

Short-term issues

Portability and compatibility for soft-copy review

1. Application of value of interest lookup tables, and preset window width and levels at diagnostic workstation

2. Chest wall justification

3. Correct image orientation

4. Standardized image labeling

5. Maintain black background upon window width and leveling

Long-term issues

Workflow: general

1. IHE scheduled workflow

2. IHE reporting

3. Structured reporting

4. IHE postprocessing workflow

a. CAD issues

b. 3D and other postprocessing

RIS/PACS

1. Modality work list

2. Graphical protocol for mammography images

3. Synchronous communication between acquisition workstation and soft-copy review

4. PACS work list availability at any soft-copy review station

5. Prefetch issues

6. Storage and archiving

Printing and removable media

1. Printing

2. Portability of mammogram

a. Format and capacity for burning to CD/DVD

b. Format and capacity for uploading images for comparison

Source: Report from the initial planning meeting of the Integrating the Healthcare Enterprise Digital Mammography Task Force, July 2005