At the University of California, Davis (UC Davis), we had the opportunity to participate in the Digital Mammographic Imaging Screening Trial (DMIST).¹ As part of this trial, we imaged approximately 3000 women using a prototype of the Fuji FCR 5000MA digital mammography system (FUJIFILM Medical Systems USA, Inc., Stamford, CT). This article reviews our 2-year experience using this system as part of this major study.

Benefits of digital mammography

Converting from screen-film to digital mammography offers several advantages. One of the most important advantages is the ability to overcome the limitations of analog mammography—specifically, the limited exposure latitude caused by the need for high contrast to detect subtle lesions in the breast. For dense, thick breasts, the transmitted exposure covers a much greater range than does the screen-film latitude, causing overexposure in thin areas of the breast near the skin line, and underexposure in the highly attenuating glandular tissues, causing the loss of anatomical detail and contrast. Digital detectors used for mammography have a wide-latitude response that can capture the X-ray information in the over-and underpenetrated regions and provide excellent contrast by digital image postprocessing enhancement methods. In addition, digital imaging provides the opportunity to advance processing techniques that may reveal additional information not seen on film; it can also easily perform a “second read” using computer-aided detection (CAD). Digital technology also facilitates remote diagnosis. With telemammography, images acquired at a remote imaging center can be transmitted electronically to a central location for diagnosis or consultation. Reduced image handling and electronic storage and retrieval also greatly improve workflow.

Digital mammography provides an extended dynamic range (latitude) as a result of the way the image information is acquired on the digital detector and converted into a digital number. Screen-film detectors have extremely narrow latitude because of the need to have high contrast (optical density differences) to obtain a small difference in transmitted exposure––these factors are related because the film is both the acquisition and display medium. Often, the densest areas of the breast image are underexposed, and the most highly transmitted areas near the skin line are overexposed. Digital systems, on the other hand, have the acquisition and display decoupled, allowing for image postprocessing for image contrast and resolution enhancement, limited only by the signal-to-noise ratio of the image itself. Thus, processing can be applied to the digital mammography image to provide image information for the densest regions of the breast while also evaluating the anatomy at the skin line and peripheral areas.

For the radiologist interpreting the images, however, digital processing initially presents challenges because of a completely different presentation and look relative to the screen-film images, which makes longitudinal comparisons initially difficult. For those converting to digital mammography, this is certainly an issue that must be dealt with carefully during the relatively long transition from analog to digital. On the other hand, because of the flexibility of postprocessing, the radiologist can develop a greater confidence in the interpretation of the difficult cases that are suboptimally presented on film.

CR versus DR

Currently, there are 2 types of digital mammography systems available in the United States: computed radiography (CR) and digital radiography (DR). A CR system was recently approved by the U.S. Food and Drug Administration (FDA) for use in breast imaging. With CR systems, the X-rays transmitted through the breast, antiscatter grid, and cassette cover are absorbed by the CR imaging plate, a photostimulable storage phosphor (PSP). Locally absorbed X-ray energy corresponding to anatomical variations in the breast produces an electronic latent image on the PSP. Subsequently, the cassette is removed from the mammography system and is placed in a CR reader with a scanning laser beam that stimulates the release of light that corresponds to the incident X-ray intensity. The light information is captured, converted to a digital signal, and displayed at the workstation (Figure 1). With DR, the X-ray signal is converted directly to a digital signal at the acquisition stand in the detector and no cassette is used. The image is displayed at the workstation shortly after it is acquired.

Using a CR system

With CR, the image acquisition pro-cess is nearly identical to that used with analog mammography. The CR cassettes are identical in size and function to screen-film cassettes (18 × 24 cm and 24 × 30 cm), and the image acquisition de-vice is set in the conventional way to the required size that best matches the breast size. This means that the technologist does not need to image a small breast on a large panel or image a large breast on a small panel and “tile” the images to obtain a complete breast examination. Once the image is taken, the cassette is removed from the acquisition stand and is placed in a digital reader and processed before the image can be viewed. Each image is then viewed by the technologist at the quality control (QC) workstation to ensure proper imaging, including appropriate positioning, lack of motion, etc.

When dedicated CR mammography was first tested in the United States under a research protocol approximately 5 years ago, the detector system was simply a high-resolution imaging plate with specialized cassettes for mammography using conventional CR readers. However, since then, CR mammography systems that are nearing market approval have improved with the introduction of finer sampling (50-µm laser spot size) and the ability to collect more light from the photostimulated luminescence (PSL) process using 2 light-channeling guides. The readout is tuned for high resolution and low noise, both of which are extremely important in digital mammography.

In clinical operation, CR mammography is very similar to screen-film mammography. One difference is the increased X-ray absorption of the CR cassette and imaging plate by 20% to 30% more compared with a screen-film detector. Acquiring images at approximately the same dose in the DMIST study required an adjustment of the automatic exposure control (AEC) sensitivity by a similar amount, as the AEC detector is positioned underneath the cassette. The solution was to use the density selector switch at the “-2” position (each position changes exposure typically by 12% to 15%) for the CR cassette, and the “0” position (as calibrated) for the screen-film cassette. For the “-2” setting, the electronics for the AEC system turned off the X-rays at the appropriate time to achieve approximately the same average glandular dose to the breast. In terms of acquisition techniques, the X-ray generator selected the “optimal” kVp and attenuation filter (either molybdenum or rhodium) using a brief test-shot method to evaluate the penetrability of the breast and algorithms tuned for screen-film response. In most cases, the techniques used for screen-film and CR cassettes were within 1 kV and ±10% of the mAs, although occasionally there was a greater difference (usually the CR system would drive the kV higher and mAs lower). Certainly, for a system tuned for dedicated digital acquisition, in all likelihood a slightly increased kVp and lower mAs could be used to reduce breast dose without a loss of image quality.

Originally, CR cassettes for mammography were designed with single-sided readout. The recent introduction of dual-sided imaging plates and reader systems allows a more efficient collection of photostimulated light from the laser beam by providing light collection from both the front and back sides of the imaging plate. Functionally, the cassette is used in the same way as a single-sided CR cassette. The imaging plate itself is composed of the PSP material layered on an optically transparent support. After exposure, the cassette is placed into the reader, the imaging plate is extracted and translated through an optical stage (Figure 2), and PSL is generated from the laser beam in both the forward and backward directions. Light collection guides are positioned above and below the imaging plate to capture and measure the light intensity, which is then amplified and converted to a digital number that is proportional to the X-rays absorbed on the plate at that position. Positional information is determined by the location of the plate in the translation stage direction and the position of the mirrors for the laser beam scan direction. There are differences in the characteristics of the information acquired from the front and back light guides. Sophisticated signal-processing algorithms are applied to the separate signals to optimize the characteristics of spatial resolution and contrast resolution, which are then combined at the image processor to produce the final output image. The read-out, which occurs as a result of the laser beam scanning the plate in raster fashion, takes approximately 60 to 75 seconds to complete. The cassette is then erased and reused.

In our experience with CR, we have found that when an area of the imaging plate is overexposed, the raw radiation on the imaging plate is recognized by the reader, and a longer erasure cycle is implemented. It is important for all residual, latent image centers to be eliminated during the erasure process. In some cases, this can take as long as the time required for readout, but it is necessary in order to avoid ghosting artifacts in subsequent images.

Spatial resolution

The effective resolution of screen-film mammography is approximately 25 µm, equivalent to 20 line pairs per mm sampling in a digital detector. To be equivalent, a digital detector for a single 18 × 24-cm image would result in 140 megabytes (MB) of data—obviously way too much. Using 50-µm pixels, about 16 million individual detector element values are output to the display, with each image made up of approximately 32 MB of data. Systems that use 100-µm pixels produce 8-MB images for an 18 × 24-cm field of view.

How does spatial resolution impact information transfer in terms of element size? If an object is larger than the detector element, a faithful representation will be obtained. On the other hand, if an object, such as a microcalcification, is smaller than the detector element, the information content will be blurred over the detector element area. The modulation transfer function (MTF) (Figure 3A) illustrates how information is lost as a function of spatial frequency (inverse of object size); a perfect system would deliver 100% modulation for all spatial frequencies. The cutoff frequency (maximum spatial frequency contained in a signal averaged over an area) for a 50-µm element size is 20 line pairs per mm. Depending on the sampling pitch (distance between sample areas), the Nyquist frequency (maximum useful frequency) when the sampling pitch equals the aperture dimension (the situation for most digital detectors) is equal to half the cutoff frequency (known as the Nyquist sampling theorem), meaning that 10 line pairs per mm is the maximum useable frequency in the acquired image for a 50-µm spot dimension.

In reality, when one compares the hypothetical perfect detector to actual CR measurements, it is clear that the MTF does fall off significantly at higher spatial frequencies (smaller object size) as shown in Figure 3B. This is chiefly because of PSL light spread during ac-quisition of the latent image in the CR reader. When considering the cutoff frequency, Nyquist frequency, and transmitted detail resolution, however, it is im- portant to consider what happens at all resolutions. The MTF shows that data. There is a loss of modulation due to light-scattering events, but high-contrast objects such as microcalcifications provide sufficient signal modulation to still be detected reasonably well with CR.

Detective quantum efficiency

Detective quantum efficiency (DQE) is the percentage of information content available to the detector that is actually used and preserved in the image, and, like MTF, is a function of spatial frequency. As shown in Table 1, when using a dual readout detector, the DQE is higher for CR than for a corresponding screen-film detector. This is because of the higher absorption efficiency of CR and a lack of grain noise, which is a problem with film. Compared with DR, CR has a lower DQE, and a slightly higher exposure is necessary to achieve the same signal-to-noise ratio in the breast image.

Image noise sources (other than X-ray quantum noise) that can decrease the DQE include luminescence noise (X-ray to light variation), pattern noise (readout, raster scan, grid signals), background noise (sensitivity, offset variation), and structure noise (detector, equipment artifacts). With CR, structure noise such as variations in the light-channeling guide response can produce a nonuniform output image (often called “shading”). Shading corrections (measuring the response with a uniform field and creating an inverse pattern that cancels the fixed patterns) will improve DQE significantly. This is implemented as a 1-dimensional correction algorithm along the path of the laser beam scan. One of the things that CR does not do, at least with current technology for breast imaging, is a 2-dimensional (2D) “flat-field” correction to compensate for consistent variations such as the heel effect, which large-area, flat-panel detectors can provide because of the fixed geometry of the source and detector positions.

With regard to data manipulation and image preprocessing, besides correction for variations in shading, for dual-sided readout the front and the back responses of the imaging plate are “weighted” to optimize image quality, considering the propagation differences of light that is transmitted versus reflected. This plays an important part in maintaining good spatial resolution and keeping the noise as low as practical.

CR versus screen-film mammography

There are several potential advantages of CR compared with screen-film mammography. One is the higher DQE associated with the digital technology that allows the CR system to provide a higher signal-to-noise ratio at similar or lower radiation doses than an analog system. Digital imaging also facilitates additional image processing and CAD evaluation. In addition, CR offers more consistent image quality with minimal artifacts (unlike the common wet processing artifacts and variation in processing chemistry). In our experience, we also found that we had fewer retakes with the CR system compared with the analog units, chiefly because of under/overexposure situations that require screen-film retake but are not a problem for the CR system because of its ability to compensate through postprocessing methods.

On the downside, the prototype CR system we used in the DMIST trial required a longer processing time than screen-film mammography, chiefly because of the single-plate reader that was used in a batch mode, requiring the technologist to insert each cassette one by one and to wait for the readout and erasure before the next imaging plate could be inserted. Also, in the trial, only film images were compared, adding another (slow) step to the process (which, of course, would be streamlined in a clinical production unit if film were required). In addition, there was some loss of detail in the smallest microcalcifications that could be appreciated on screen-film that were often not clearly seen on CR images because of the lack of signal modulation at the intermediate-to-high spatial frequencies. Finally, not unlike any other digital system, the use of CR also requires the radiologist to become familiar with the image characteristics of digital mammography.

CR versus DR

When comparing CR with DR, there are also some potential advantages to CR. First, CR is less expensive to implement. Existing mammography systems can be converted to digital with just the addition of CR cassettes and a CR reader tuned for mammography. In addition, with the purchase of one high-throughput centralized reader, facilities can convert several mammography rooms to digital technology without replacing substantial amounts of equipment. Computed radiography also offers 2 detector sizes for optimal positioning of small or large breasts, whereas DR offers only a single detector size that often compromises positioning or requires a tile-mode acquisition. The acquisition process of CR is very similar to that of analog mammography, so the technicians and radiologists do not need to learn new acquisition techniques.

The disadvantages of CR as compared with DR include the need for significant handling of CR plates/cassettes and the delay in image display as the plates are processed. In contrast, DR systems produce an image within seconds, which allows the technologist to immediately perform quality control on the image for positioning, motion, and other issues, and if necessary, perform a retake before moving to the next projection. Technique information from the X-ray generator (kVp/mAs/focal spot size, tube target, tube filter, acquisition algorithm, AEC “density” setting, etc.) and peripheral devices (compression thickness, use of grid, AEC detector position, etc.) require a modification/interface to the mammography system to download information to the DICOM header of the digital mammography image prior to sending it to the dedicated mammography workstation or universal DICOM-compliant picture archiving and communication system (PACS). Another potential disadvantage is the lower signal-to-noise ratio for the same breast dose because of lower DQE and the slightly lower intrinsic resolution of the CR system relative to the flat-panel detectors designed for mammography.

Workflow also can be a concern with the CR system because of the processing requirements; historically, all 4 screening views were acquired and the technologist left the room to process the images. One way to address this issue is to have an “in-room single-plate reader.” This provides the ability to process one view while setting up for the next view so that at the end of the examination, there will be only one imaging plate left to read out. This can enhance the throughput and make it easier to keep up with the workflow in a busy room. The downside is the need to buy an in-room reader for each digital system, with increased costs for implementation compared with a single high-throughput stacker in a facility with multiple digital mammography rooms.

Looking forward

At UC Davis, our experience with the prototype CR mammography unit was mainly positive, based on the quality of the digital images, the ease of implementing the CR detectors with our existing mammography systems, and the fact that we did not experience any technical problems with the CR reader in performing approximately 3000 studies (including a lot of QC testing) in >2 years of operation. We would have preferred not having to print film (but that was part of the study protocol), and the slowness of the prototype system (at least in batch-mode processing) was of concern.

As digital mammography continues to evolve, there are several enhancements that could be made to increase the efficiency of CR systems. The addition of an in-room single-plate reader would boost workflow by allowing the technologist to remain in the room during the entire examination. Alternatively, a batch-mode process would certainly indicate the need for a multiplate reader system. Workflow could also be enhanced by the addition of an equipment interface for X-ray technique information. Finally, the adoption of normalized, linear “for processing” image data standards would allow users to compare all digital systems equitably.

The conclusion of the DMIST study unequivocally supported the superiority of digital mammography over screen-film¹ and portends the adoption of digital mammography in lieu of screen-film with increasing frequency. This increasing implemenation will continue despite some of the hurdles that must be overcome in transitioning from an analog to digital environment, including cost and reimbursement issues, hybrid digital and analog reading for a period of time, workflow optimization, and image appearance differences among the digital mammography systems. At the time of this writing, the Fuji CR mammography system was just FDA-approved for clinical imaging. Without a doubt, the system will fill a large niche in the digital mammography market, as there is a demand for a capable, cost-effective technology that delivers excellent image quality.

Acknowledgments

The author wishes to acknowledge FUJIFILM Medical Systems USA, Inc. for providing the digital mammography system used at UC Davis for the DMIST study and for information on workflow. This article was supported, in part, by the ACRIN DMIST grant at UC Davis (Karen Lindfors, MD, Principal Investigator at UC Davis). The author also thanks Martin Yaffe, PhD, Physics Core leader, DMIST study and implementer of QC phantom tools/ software.

REFERENCE

1. Pisano ED, Gatsonis C, Hendrick E, et al. Diagnostic performance of digital versus film mammography for breast-cancer screening. N Engl J Med. 2005;353:1773-1783.