The first digital mammography (DM) systems were approved for use in the United States in 2000. At press time, five systems from three manufacturers were available for purchase in the United States, and several others have either been discontinued or are sold only in other countries (Table 1). The capital cost of DM equipment is considerably higher than that of screen-film systems. This factor, the uncertainty as to the clinical or other advantages of DM, and the concern about buying a system that might quickly become obsolete have resulted in the relatively slow uptake of this technology. The recent publication of the results of the American College of Radiology Imaging Network (ACRIN of Philadelphia) Digital Mammography Imaging Screening Trial (DMIST) indicated superior performance of DM in subgroups of the study population: dense breasts and women younger than 50. In those women, DM was seen to be more sensitive in detecting cancer with no increase in the false-positive rate compared to film. This finding has restimulated interest in acquiring DM systems. Another factor is that in many departments, mammography is the last holdout of analog technology, and facilities have a strong incentive to be completely digital and eliminate the costs and inefficiencies associated with chemical processing and archiving of films.
Many departments considering the purchase of DM are confused as to which technology to purchase. This article will not answer that question; the simple answer is that all of the current technologies have pluses and minuses associated with them. Compared to screen-film mammography, there are major fundamental differences in the technologies available. Instead, this article will clarify these differences and some of their implications, with the goal of helping to facilitate your facility’s decision. The best decision will depend on how you weigh these differences in your department. The four factors to consider in a system include imaging geometry, detectors, image processing, and display systems.
Whatever system is purchased, it is essential that technical personnel at your facility work closely with the engineers from the vendor to ensure that the site is properly prepared for the equipment. This includes properly conditioned electrical power, ventilation and air conditioning, dust control, and proper lighting conditions. Generally, an Internet connection is required, but is desirable in any case, as modern systems often can be monitored, faults can be diagnosed, and software can be updated remotely.
There are two major acquisition geometries: snapshot and scanning. At the time of writing, only snapshot systems are available in the United States as new systems; one scanning system approved by the US Food and Drug Administration is in use, but it is no longer manufactured.
Snapshot systems acquire the image using a full-area detector and a single, brief x-ray exposure. The advantages of these systems include the short exposure time resulting in freedom from motion blur or registration artifacts. Acquisition time can be important for procedures that involve a rapid series of images (for example, contrast subtraction imaging). Scanning systems use detectors that move across the breast in synchrony with one or more slit- or slot-shaped x-ray beams. Although these systems typically take a few seconds to acquire the image, they do not require an antiscatter grid, which generally provides a dose-reduction advantage compared to the snapshot systems.
Other key geometric considerations are the detector size(s), the thickness of the detector assembly, and the ability to image close to the chest wall. For smaller detectors, it frequently might be necessary to make several exposures to cover a single large breast. If exposed regions overlap, the breast dose will increase. In addition, the radiologist will be faced with manipulating and interpreting more images. If the detector assembly is too large or too thick, it could be more difficult to position and obtain optimal imaging of some women. It is always important that the detector has as minimal a “dead area” as possible on the edge proximal to the patient’s chest wall so as little tissue as possible will be excluded from the mammogram.
The detector is the heart of the DM system. Systems can be differentiated in terms of their generic type of detector and image-acquisition geometry as flat-panel systems, scanning systems, and photostimulable phosphor (PSP) systems. Many people refer to PSP systems as computed radiography (CR); however, this is actually one manufacturer’s name for its technology, and the term is confusing because all DM systems use computed radiography. In the PSP systems used for mammography, the x-ray?sensitive phosphor plates are held in cassettes that must be inserted in the x-ray unit for exposure and then moved to a separate image-reading unit. This is both a strength and a disadvantage. On one hand, it allows DM to be performed in an essentially conventional x-ray mammography unit; multiple units can share a single reader. This ability reduces capital cost significantly. At the same time, the need to handle cassettes is labor-intensive and generally reduces productivity of the technologist.
Flat-panel systems (Figure 1) get their name from the flat-panel display technology used as their readout mechanism. This technology is essentially the same technology that is used for flat-screen televisions and laptop computers; however, with flat-panel displays, it is used in reverse, to acquire the image rather than display it. It consists of a large array of tiny elements fabricated on a glass plate. Each detector element (del) is equipped with a microscopic thin-film transistor (TFT) switch that is used to direct the image signal acquired on that element to a readout line connected to a digitizer. The del size defines the maximum spatial resolution of the detector. But factors other than the del size also can affect resolution.
|Figure 1. System performance begins with the effective size of the del and the space between dels, or pitch.|
Currently on the market are two major types of flat-panel detectors whose basic operation is quite different: phosphor and direct-conversion detectors. In phosphor systems, the x-rays are absorbed by a cesium iodide (CsI) phosphor, and the light produced is collected by photodiodes located on each element. The photodiodes produce an electronic charge proportional to the level of x-rays transmitted by the breast, and the charge is read out from the array and digitized.
In the direct-conversion detector, the x-rays are absorbed by an amorphous selenium layer and converted directly to charge, which is collected by electrode elements on the plate and read out in a similar manner.
One important consideration is the rate at which images can be acquired. Some systems require an extensive detector-preparation cycle between images. The required interval is likely to change as technology evolves, so it is important—especially in high-volume facilities—to establish from the vendor what the interimage time is.
System performance starts with the effective size of the del and the spacing between dels, also called the pitch. The effective size or aperture can be smaller than the pitch if part of the del is insensitive to x-rays. In the case of the flat-panels, this is the “real estate” occupied by the switches and the readout lines. A smaller aperture causes the image to be sharper, but it also can cause a reduction in detector sensitivity and information to be missed. When the aperture is smaller than the pitch, a phenomenon called signal aliasing is more likely to occur. Aliasing causes information to be incorrectly rendered in the image, both suppressing some spatial frequencies and giving the impression of signal information that does not actually exist. Noise aliasing is a similar process that causes an increase in the apparent image noise. In most current DM systems, signal aliasing is not clinically apparent, but noise aliasing may be measurable.
The effective aperture also can be larger than the pitch. This can be due to blurring by spread of light in a CsI phosphor or, in the case of PSP systems, scattering of the readout laser light in the photostimulable-phosphor screen. In this case, the image may be less sharp; however, the blurring can reduce aliasing. This phenomenon may be important in considering differences in performance between direct-conversion and phosphor-based systems. In the former, the effective del aperture is more likely to be close to the pitch, giving rise to an inherently sharper image with more aliasing; whereas in the latter, the larger effective aperture caused by slight blurring may result in the opposite being true, so there may be a trade-off between sharpness and noise.
The DICOM standard provides for two levels of digital mammograms: the “for processing” image which has had only fairly simple operations performed on the information produced by the detector, and the “for presentation” image which may have been subjected to more extensive image processing.
Image processing is an important feature of all DM systems. The initial processing is generally a “flat-field” or gain correction, where spatial nonuniformity in the detector sensitivity may be corrected by imaging a uniformly attenuating object and creating a gain map that can be used to correct all subsequently acquired images. For flat-panel systems and scanning systems, this transformation also corrects for nonuniformities in the x-ray field—an example is the heel effect. For current PSP systems, the correction is applied only to the laser readout system and not to the phosphor plates themselves or the x-ray field. Therefore, residual artifacts occasionally can be observed.
Other image processing differs among manufacturers and may include:
- Peripheral compensation to flatten the signal level at the edge of the breast. This essentially suppresses the effect of thickness reduction near the edge and reduces the dynamic range of image signal that the display system must accommodate, allowing higher contrast settings to be used in image display. It is important that when such software is used, it does not unduly distort the contour of the breast.
- It is possible to correct to some extent for detector blurring through image processing by deconvolving the blurring function of the detector. This procedure can be very effective, but if overdone, it also will enhance image noise. Therefore, it is important that the inherent detector resolution be adequate and that the image noise level is acceptable. The latter is accomplished, in part, through careful design of low-noise detectors as well as through appropriate design and use of automatic exposure control and/or automatic technique control.
- Other image enhancements also are employed to attempt to optimize contrast through the breast and best use the limited dynamic range of the display system.
Image-enhancement techniques are proprietary to each vendor and frequently can be applied or not at the user’s discretion. The best way to evaluate these algorithms is to observe the rendition of key structures: spiculations, microcalcifications, and margins of benign and malignant lesions with and without the enhancement activated in a series of sample cases, including images of both dense and fatty breasts.
Image processing can be performed at the acquisition system or at the display workstation and where this is done differs among systems. In the future, as third-party interpretation workstations become more widely used, it is likely that most of the processing will be carried out at the acquisition system and the “for presentation” images will be sent to the interpretation workstation or PACS. If considerable processing has been done on an image to improve its visual appearance for interpretation, some of the quantitative information relating directly to x-ray transmission of the breast may be lost. If such information is required for specific applications, it will necessary to save images in the “for processing” format.
The display system plays a major role in influencing overall performance of the DM unit, both in terms of the ease of image interpretation and the image quality presented to the radiologist. Although some radiologists use hard-copy systems (laser-printed films) for interpretation, in the long term, the benefits of DM and cost efficiency will be fully realized only if soft-copy display is used. Flat-panel LCDs are more compact and produce far less heat than the conventional cathode ray tube (CRT); however, they have a more limited viewing angle than CRTs.
The display must have a suitable number of high-quality monitors—normally, two 5- megapixel (MP) monitors are recommended—to allow viewing of as much of the mammogram as possible at the required resolution level. Remember that a 5-MP monitor is capable of displaying only a single mammogram with 100-mm dels at full resolution. If multiple images or images with smaller dels are displayed simultaneously, as is normally the case in mammography, then they must be viewed at reduced resolution and then the images panned and zoomed to inspect structures of interest at full resolution.
The monitor on the acquisition station often is overlooked. The quality must be high enough to allow the technologist to assess the adequacy of the acquired image without having to walk to the radiologist’s workstation, which may be located a considerable distance away. If needle localizations are to be performed on the system, the image quality on this monitor and image-manipulation operations must be adequate to provide the required image quality.
Display software varies greatly among system types and is a major factor that determines user satisfaction with the DM system. Space constraints here prohibit reviewing all of the issues, but some important questions to ask are:
- How convenient are basic image-manipulation operations, those that will be used with essentially every image?
- Is there flexibility of image-hanging protocols?
- Can the system handle images acquired on another vendor’s system and display them acceptably well?
These issues will take on great importance as DM becomes more widely accepted and facilities purchase multiple acquisition units. There is a natural and understandable tension between the proprietary interests of manufacturers in protecting their intellectual property and the need for different systems to work seamlessly together. To address these issues, the Integrating the Healthcare Enterprise (IHE) group is developing guidelines and standards in the form of an “Integration Profile” for image display and intersystem compatibility. As this is a rapidly moving area, it would be a good idea to keep up with the ongoing activity via IHE’s Web site (www.ihe.net).
Martin J. Yaffe, PhD, is senior scientist of the Imaging Research Program at the Sunnybrook Health Sciences Centre at the University of Toronto.