Digital mammography

Digital images are very data-intensive. A single 18 × 24-cm breast image acquired on a full-field digital mammography (FFDM) system with a 50-µm detector will contain approximately 35 MB of data. Systems with smaller pixel-size detectors produce significantly more data (Table 1). Multiply that by 4 or 6 or more images per study and by 10,000 studies per year, and you are left with a large amount of data (several terabytes [TB] ) that must be networked, stored, easily accessed, and optimally viewed. There are certainly practices that read ≥50,000 mammography studies per year. How many years of comparison studies do we need? Five? Ten? This means that we may all need to start learning about a new unit of data: the petabyte (10¹⁵). I studied Greek as well as Latin, but I had to use Google to find out what that level of storage would be called. One of the issues that radiologists face is that many of the tools we use are relatively primitive when it comes to dealing with such large amounts of data. Currently, state-of-the-art diagnostic display monitors, for example, provide 5 megapixels (MP) of data. When looking at a standard 4-view display of such mammographic images on a 5-MP monitor, the radiologist is seeing only 1/16 of the information. In the near future, images may be available that contain 140 MB of data each. How will clinicians view these?

The current trend is to make digital resolution as close to film as possible. While contrast resolution has been the strong point of digital mammography, there have been doubts about giving up the previously unrivaled high spatial resolution of analog mammography, with its depiction of fine calcifications and thin spiculations that allow us to find and diagnose breast cancers. So improved digital resolution will continue to be a part of the digital revolution, in much the same way as we are seeing this in the quest to change a century-and-a-half of photographic imaging, with digital single-lens reflex cameras that now can have up to 17-MP image receptors and 8-gigabyte (GB) compact flash storage. The question is, however, whether we as radiologists are truly satisfied with the resolution of film mammography. When viewing microcalcifications on a mammogram, the radiologist is actually seeing an aggregate 50-to-several-hundred µm in size, not the individual microcalcifications, because both analog and current digital systems are inadequate to clearly image objects this small.

I once conducted a small study in which I asked the pathologist to measure the actual calcifications that were reported. Generally speaking, the benign calcifications were roughly 10 µm in size. The malignant calcifications were approximately 100 µm to 500 µm. Clinical phantom testing done as a part of the Illinois Radiological Society mammography accreditation program in the late 1980s, (which was an important precursor of the American College of Radiology Mammography Accreditation Program) showed that it is virtually impossible to routinely see less than a 100-µm calcification on screen-film mammography. This means that there are going to be calcifications that are not visible on analog mammography. With improved technology, however, we may one day be able to image even the tiniest calcifications and, thereby, improve our diagnoses and detect earlier cancers.

Networking and standards

Digital Imaging and Communications in Medicine (DICOM) standards are used in all facets of radiology today. The specific standards for digital mammography, however, are still evolving. Many equipment manufacturers still use proprietary subfields that make it difficult to transmit images across varying systems. Eventually, all manufacturers will need to embrace a single standard so that all images will be viewable on all workstations and so that all processing and manipulation of the images can be performed in the same manner regardless of the manufacturer. Since different image-processing algorithms are now evolving, the question of whether raw image data storage should routinely be done is raised. Processed image storage, which is what is most frequently done now, may cause problems in the future when algorithms and display methods change, if the new methods cannot be used to make the older images similar in appearance for comparison purposes. As I point out time and again in my mammography reports when I have inadequate (or no) comparison films, mammography works best by looking for changes. So our sensitivity is dependent on our ability to tell whether a change in appearance is due to a change in processing for presentation or to a real change in the tissues. If you haven’t yet done this, compare several currently available systems; the differences can be surprising. Mammograms obtained on digital systems from Hologic, Inc. (Bedford, MA), Fischer Imaging Corporation (Northglenn, CO), and GE Healthcare (Chalfont St. Giles, UK) all have distinctly different “looks” that can sometimes confuse even an experienced observer when direct comparisons are made.

Another standard that needs to be developed is one that will allow the user to annotate images. With film, notes can be written directly on the image using a crayon. How do we do this with digital images? Manufacturers need to develop conventions for annotations. Again, this must be standardized. Will the annotations be permanent or erasable? Who will have the ability to annotate images: the technologists, the radiologist, the clinician, the paralegal in the plaintiff’s law firm? All of these issues must be addressed.

Data storage

Physical storage of films used to be a problem. This is one of the areas in which digital technology has greatly improved the process. As noted above, if you have roughly 35 MB of data per image (small image receptor, 50-µm pixels) and you acquire 6 images per patient, and you have 10,000 patients, you will need 2.1 TB of storage per year, or, over a 10-year period, approximately 20 TB of storage. Larger patient volumes at some institutions could push that requirement upwards of ≥100 TB. Fortunately, the cost of electronic storage has been dropping rapidly as technology improves, so cost is generally not the most significant issue. But the speed of retrieval is.

There are other storage issues that need to be addressed, however. Images acquired using larger field sizes or higher resolution detectors produce more data, and, therefore, the storage requirements increase. Saving both the raw and processed data will nearly double the amount of storage needed.

These storage requirements can be somewhat decreased by data compression. There are several options for compression, including lossless and lossy compression from which facilities can choose. While this is not my area of expertise, I believe significant decreases in data storage requirements are likely to be achieved for processed images, as has been shown for chest imaging. But that specific research is still in progress for the detail-dependent modality of mammography. Since the future is not yet here though, what if computer-aided detection (CAD) temporal subtraction techniques evolve that work best on uncompressed raw data? Academic institutions may be more inclined to hedge their bets in this area, and opt for storing the raw data.

Local image storage is another concern that must be addressed. Many workstations hold very little data, usually 60 to 80 GB—less hard drive storage than is common on today’s laptops (the portable computer I am writing this on has 100 GB storage, and my desktop Macintosh G5 has 500 GB). In most facilities, this is only approximately 1 month’s worth of images. Therefore, the retrieval of images for patients who were last imaged >1 month earlier may be delayed while the images are retrieved from the PACS, particularly if this is a manual push initiated by the technologist or the time-frustrated radiologist in a diagnostic mammography setting. Increased short-term storage capabilities could greatly increase clinical efficiency. As we get more digital workstations in mammography, it will become increasingly important to have scheduled patients’ previous images (which eventually will be multiple examinations over years) prefetched by PACS worklists to the correct workstation.

Image movement

Image movement is still an issue with digital mammography. With early digital mammography systems, the acquisition workstation was connected directly to the review workstation and all images were contained within the local system. This meant that all image reading had to be conducted on the system’s dedicated workstation. This is not efficient. We must have vendor-neutral, multimodality workstations so that all images can be viewed on all workstations regardless of the manufacturer.It is very difficult to share a workstation when reading imaging studies, and, therefore, it is best to have 1 review station for each full-time equivalent radiologist working each day.

In addition, images should be sent through a quality control workstation before the data are entered into the PACS. Because the patient data is burned into the examination, it is essential that the information be correct. My personal record in the analog era at the University of Chicago is one 4-view study with 3 different patient names on it, but I suspect that there probably is a record of 4 names on the 4 views somewhere. The bottom line is, however, that once an error is entered into the digital data and the data is sent to the PACS, there is virtually no way to retrieve the images again. Most current systems do not allow users to easily change the patient’s name or other data, and, unlike film, you cannot just put a sticker on an electronic image. This is another area for which standards must be developed.

Retrieval speed can be an issue for digital mammography workflow as well, depending on the background network being used. As we have seen, as the image volume in breast MR imaging has increased, the switch capacity, network node distribution, and PACS priority categorization of different image modalities can all affect our ability to view the images in a timely fashion.

Auto-push to the PACS is desirable to make images widely available after acquisition and to ensure their storage, but even more desirable is auto-pull from the daily worklist. With a worklist-driven PACS, auto-pull can greatly improve workflow. Inputting the patient information by hand in order to locate studies can be very time-consuming. Also, if images are sent directly from the acquisition workstation to the review workstation and old images have not been retrieved from the PACS to the review workstation before the new images arrive, the workstation may not recognize this as the same patient, resulting in different examinations for 1 patient appearing twice on the worklist, as if they were 2 different patients. This makes comparison very difficult on our current workstation, and this was not a rare occurrence before we worked to correct it.

Right now, each of the digital mammography systems available in the United States uses different detector technology and different processing. We have already seen one of these systems pass into what is essentially obsolescence: the Fischer SenoScan will no longer be manufactured after acquisition of the intellectual property rights by Hologic earlier this year. This situation is not likely to be unique and emphasizes that flexibility in image retrieval and display is important, as the proprietary Fischer workstation that is part of our unit does not allow viewing, for example, of GE images. So it clearly has a limited lifetime now that this is a “legacy product.”

In the earlier phases of digital mammography development, sticking with one manufacturer could shield you a bit from some of these soft-copy incompatibility concerns, but as digital mammography increases over the next few years, it is clear, certainly at a tertiary referral hospital like mine, that we will have to be able to view images from other institutions that have made different equipment purchase decisions. So, the era of the proprietary workstations is appropriately ending, and these will soon be obsolete. The proliferation of ancillary workstations to do certain image processing and CAD is also an area that has not yet been clarified. Software from third-party sources and integration of all the modalities that are important to breast imaging will have to be more accessible on the primary workstations of the future, if we are not to drown in too many monitors.

Putting it all together

Who is really responsible for making this system as efficient as possible? Some may say it is the digital mammography vendor’s problem. Others say it is the PACS provider’s responsibility. Still others say it is the workstation manufacturer’s problem. But the truth is, as the radiologist, it is your problem. Therefore, when purchasing equipment, it is imperative that clear performance standards be set with the vendor. Clearly state in all contracts that systems to be purchased must work within certain specifications in your information technology (IT)/PACS environment or they will be removed. When shopping for a new system, ask to be shown another facility where the system is in use or ask for some form of performance guarantee. Pretend to be from Missouri, the “Show me” state.

Problem solving

At the University of Chicago, we have been using digital mammography in conjunction with our PACS since 2002, during which time the PACS has changed once. Neither PACS has been intrinsically friendly to digital mammography, as the PACS vendors have not yet had significant demands for this type of product. During this time, several issues have arisen that we have tried to address. One issue was the time-consuming problem of having to manually push or pull the digital mammograms to and from the PACS. I thought we would be able to eliminate this issue with the installation of our new PACS (since it was supposed to automate these functions), but as I write this, it has become an increasingly time-consuming task for the technologists to continue to do this manually. For a low-volume, single-site operation, this may not be a problem, but for our facility this is still a thorn in our digital sides that needs removal.

Another problem was that the digital mammograms acquired on one system could not be viewed on a different vendor’s workstation. This is in the process of being fixed by the purchase of a multivendor, multimodality workstation, but this type of workstation is still in its developmental stages. We also became aware that we were unable to use the Premium View features (GE Healthcare, Chalfont St. Giles, UK) on stored processed images. In order to overcome this particular limitation, we wanted to begin saving the raw data as well as the processed images, but this has left us stuck between our IT team and several vendors, and this also remains unresolved.

We had difficulty retrieving old digital images for comparison with our first PACS. At first, we printed all images, but now the technologists push the images to the review workstation each day. This is not a 100% fix, however. Auto-pull would be a significantly better solution.

We also found that our CAD data was not being displayed on one particular vendor’s system, although the CAD vendor was providing compatible output. After an extended attempt by the local IT team to solve this problem, the FFDM vendor was called and was able to quickly fix the problem. When it comes to troubleshooting and servicing, it is necessary to be persistent and willing to try alternate pathways. We also found that CAD marks could not be stored in the PACS for our digital images, and neither the digitized analog images nor the CAD marks for them could be stored. As a temporizing measure, we now use an inexpensive thermal paper hard-copy print that works very nicely, and this stores CAD marks in the patient’s jacket, just as we used to do with everything before we went digital. This provides a permanent record and has been extremely beneficial many times when the electronic display is incorrect (when displayed patient information is missing or wrong, when there are system electronic display problems, etc.). While there is debate about whether CAD marks should be stored, my strong advice is to do so, as they are really part of the patient’s medical record. They can vary when analog images are redigitized or algorithms improve, and they are what you used to make your decision. But this is not the digital future. So we have now ordered software that will allow us to store the digitized images. But we are still having a problem with the overlays in the PACS for the digitally acquired mammograms, and we continue to print those results. Time will heal this too.

Breast imaging is no longer limited to mammography, and workstations of the future must accommodate multimodality reading. Currently, MR image viewing is limited to several dedicated MR workstations at our facility. We can review MR on the PACS, but we cannot do simultaneous multiplanar (axial, sagittal, and coronal) review, or get maximum intensity projections or gadolinium-uptake curves. To see these views, we have to use a special workstation or purchase special integrated software. Even with third-party software that solves some of these problems, we all prefer to use the dedicated MR workstation for the multiplanar review software that is not available on the PACS. In addition, MR images can be slow to arrive at the review workstation. In my experience with a new magnet and breast coil, they have taken up to 18 minutes just to load, not including the time spent reading the images. This is a PACS network problem, but it has been nettlesome to eliminate it. One way to make this process more efficient is to send the images directly to the workstation (direct wiring) as well as to the PACS simultaneously.

Other improvements that I would like to see in diagnostic workstations include larger viewing areas, improved resolution, the flexibility of black-and-white or color image display, voice-activated commands, and more local storage capabilities. I would also like to see the development of software designed to guide presentation, which could reside on the workstation itself. In other words, we need to develop viewing protocols as well as hanging protocols. Remember that for 50-µm digital mammography images, the current standard of 5-MP monitors displays only one quarter of the information, averaged to fit to the monitor, even when you look at only 1 image view per monitor. If you have 2 views per monitor or include comparisons on the same monitor, the displayed information content is averaged accordingly and is even less. So if you prefer to see each breast quadrant at full resolution and presented in a certain order, this could be preset at your preferred pace in the digital future. Additional automations, such as display sequences of the system, and increased CAD-like capabilities, such as progressive masking, would also be beneficial.

Conclusion

Digital mammography and the increased use of other digital breast imaging modalities promise to radically alter the way we practice. More research-stage advances, such as digital tomosynthesis and whole-breast ultrasound, hold promise for putting all the modalities that we use into 3D, which is highly desirable. To say that we have achieved great progress is true, but the challenge of implementing this progress in routine clinical practice remains, and it is likely that the pace of change will continue for some years to come. Imaging continues to increase in importance in finding, diagnosing, and monitoring treatment of breast cancer. Our medical colleagues have become increasingly dependent on our ability as radiologists to provide this information in digestible form both accurately and rapidly. The volumes of information to be digested keep increasing, leading to the sorts of hiccups along the way to the digital future that I have described above. But you cannot just hold your breath in the hope that they will go away. To get to that future requires planning, patience, and the increased insistence that vendors help us address these issues on the path of continuous quality improvement.

In the decidedly analog Hell of Dante, he would have had to struggle to come up with a Latin phrase to describe our current state of information technology (maybe just the Latin “id” for “IT” would do) as the premise that promises to either free or totally consume the world. But being a devil is a fairly mundane and predestined job, with little hope for advancement. Putting on our collective thinking caps and rolling up our sleeves can get us out of the pit, off the pyre, and into the promised land heralded by the digital enthusiasts. And we must get there. It is really our only salvation.