Log Transmitted X-Ray Intensity
Fig. 3.21. Characteristic curve for a mammographic screen-film combination.
video monitor or print it on laser film. The transformation (inset to Figure 3.22) can be readily adjusted by the user to optimize the presentation of relevant anatomical features in the breast (Yaffe, 1992).
Digital imaging detector element (del) sizes must be adequately small if fine detail in the breast is to be depicted accurately. If it is too large, then the image will be unsharp and the borders of structures will be jagged and poorly defined. Under these circumstances, while the presence of microcalcifications might be evident, details of their shape and edge structure might be inadequate.
The limiting high-contrast resolution of the screen-film image receptor for mammography is on the order of 20 line-pairs per millimeter (lp mm1). In a digital system, to obtain such resolution, the del would have to be spaced 25 pm apart or less. For a 24 x 30 cm image field, a matrix of 9,600 x 12,000 del would be required.
In practice, screen-film mammography does not resolve 20 lp mm1, because factors such as the x-ray tube focal-spot size, noise, and inherent low contrast of the image features become limiting factors. In fact, using contrast-detail test objects (Nishikawa et al., 1987), it was demonstrated that for subtle soft tissue-like structures, a digital imaging system with modest (10 lp mm1) limiting resolution could display lower contrast and smaller objects
than a state-of-the-art mammography screen-film system. This appears to be supported by early clinical experience with digital mammography systems operating at only 5 lp mm1 limiting resolution (Freedman et al., 1995), although the findings are not conclusive (Lewin et al., 2001). Freedman et al. suggested, however, that although adequate detection of structures may be achieved at 100 pm, smaller del are probably required for shape determination of these structures, often an important feature in the diagnosis of microcalcifications.
In addition to the del determined by the detector, it is also important that the transmitted x-ray intensity be measured to appropriate precision. This is determined, in part, by the number of gray levels of digitization (i.e., the number of bits in the analog to digital converter). Use of too few gray levels will cause information to be lost and will give the image a "terraced" appearance with artificial contrast that may be disturbing to the radiologist. For dig ital mammography, it appears that between 12 and 13 bit precision is required to accommodate the range of x-ray intensities adequately, unless a logarithmic analog-to-digital converter is employed, in which case, fewer bits are required.
Detectors for digital mammography should have the following characteristics: (1) efficient absorption of the incident radiation, (2) linear response over a wide range of incident intensity, (3) low intrinsic noise, (4) spatial resolution on the order of 5 to 10 cycles mm1 (50 to 100 pm sampling), (5) at least an 18 x 24 cm field size and preferably able to handle a 24 x 30 cm field size, and (6) acceptable imaging time and heat loading of the x-ray tube.
220.127.116.11 Area Detectors—Full Field. Conventional screen-film mammograms are produced with a single, brief radiation exposure of an area detector. This approach is convenient, allows good throughput, and makes efficient use of the heat loading applied to the x-ray tube. For digital mammography, the area detector must have appropriate spatial resolution, field coverage, and signal-to-noise performance. Some approaches to area detectors, their strengths and weaknesses are described below.
• Digitization of Film Mammograms: Conventional film mam-mograms can be digitized with a high-resolution optical scanner. This allows the image to be acquired quickly, although film processing and digitization require several minutes. The digitized image can then be manipulated to improve display contrast characteristics.
The quality of the digital image will be limited both by the performance of the digitizer and by the quality of the information initially stored on the film. If conventional mammo-graphic film has been used, then the main limitation in image quality will be associated with the granularity of the film emulsion. This will affect the image most at high spatial frequencies, where the modulation of image information is low compared to the noise. Attempting to achieve a large degree of contrast enhancement, in either the "toe" or "shoulder" regions of the film's response curve, may cause noise to be amplified to an unacceptable degree. Commercial digitizers typically have reduced performance at high optical densities where their system noise becomes a limiting factor in measuring the low levels of transmitted light.
Because it would require a film to be produced, processed and then digitized with the final processed digital image possibly presented on a second film; it is unlikely that this approach would be acceptable for clinical practice.
• Demagnification Cameras: Demagnification cameras for digital mammography are produced by coupling an x-ray absorbing phosphor to a smaller-area photodetector such as, a charge coupled device (CCD) array via demagnifying lenses or fiber-optic tapers. The photodetector output can then be digitized to produce a high-resolution digital image.
Such systems are employed for producing small-area (5x5 cm) digital images (Karellas et al., 1990) for guiding sterotactic breast biopsy and, typically, provide one million individual images (1,000 x 1,000) with 50 pm del. It is not practical to extend this approach to full breast imaging by employing a larger phosphor surface and increasing the amount of optical demagnification to a factor of about eight. This is very inefficient and causes image noise to be increased to unacceptable levels. On the other hand, a mosaic of multiple small-format detectors, using optics with more modest demagnification factors and acceptable efficiency can be combined to obtain a camera which can cover the full breast. We will refer to this as a Type 1 detector (Figure 3.23). It is important that the subimages from these detectors be combined (stitched) seamlessly to form the complete image so that disturbing artifacts are not introduced at the borders. One manufacturer (Lorad Corporation, Dan-bury, Connecticut) has received regulatory approval from FDA to market a system based on an array of 3 x 4 CCDs coupled by 12 fiber-optic tapers to a full-area phosphor screen.
• Photostimulable Phosphors: Photostimulable phosphors have been successfully developed as an imaging system for general radiography (Kato, 1994), and it is possible to extract the information from such devices in digital form. Energy from absorbed x rays causes electrons in the phosphor to be excited. Rather than decaying immediately to give off light, the electrons are captured and stored in traps in the phosphor crystals. The number of traps filled is proportional to the exposure received by the phosphor. The image is created by scanning the phosphor plate with a finely-focused laser beam. This stimulation releases electrons from the traps, giving rise to emission of light of a shorter wave length (blue), which is collected point-by-point
as the laser scans over the plate, as illustrated by the Type 2 system in Figure 3.23. A system with 50 pm del has been introduced. The actual spatial resolution of this technology may be determined by the scattering of laser light within the volume of the phosphor, stimulating a larger region of the material than the initial width of the laser beam. A second important factor is that the collection of stimulated light is inefficient, resulting in a loss of signal-to-noise ratio because of a secondary quantum sink in the system (Nishikawa and Yaffe, 1990). This limitation may be offset in part by reading the emitted light from both sides of the phosphor plate. Some researchers have reported positive impressions of the clinical performance of this technology (Freedman et al., 1995) although others have found that the performance is inferior to screen-film technology (Kheddache et al., 1999).
• Amorphous Silicon: Amorphous silicon provides another means for producing area detectors suitable for digital mammography. An array of light sensitive diodes is deposited on a plate of amorphous silicon such that each element provides the signal for one pixel of the image (Type 3 in Figure 3.23). The diodes are covered by a suitable x-ray absorbing phosphor such as thallium-activated cesium iodide. The electric charge stored on the capacitance of each diode after x-ray exposure can be read out through a network of switches (Antonuk et al., 1992). Challenges with this technology involve the large number of del in the receptor and the complexity of connecting read-out wires to all of the rows and columns of the matrix, while maintaining minimal loss of coverage at the chest-wall side of the imaging system. A system of this design developed by General Electric Medical Systems (Wankesha, Wisconsin) has received FDA approval for clinical use. In this system, the detector is composed of del that are approximately 100 pm on a side. The detector array resides in a bucky assembly which contains a moving grid.
18.104.22.168 Scanned-Beam Detectors. Another way to produce a high-quality mammogram is to use a small-area long, narrow (slot) detector, which is scanned in synchrony with the radiation beam, across the entire breast to build up a full image (Type 4 in Figure 3.23). In this way, images with high spatial resolution, dynamic range, and signal-to-noise ratio (SNR) can be produced.
Because the image is acquired sequentially in a scanning system, the acquisition time is longer than for an area detector. A major offsetting advantage of scanned beam systems, however, is that because only part of the volume of the breast is irradiated at any one time; it is much easier and more efficient than in an area system to control the detrimental effects of scattered radiation at the image receptor. Less scattered radiation is created during the time when the detector is measuring x rays from a particular part of the breast. In fact, the scattered radiation contribution to the detected signal is sufficiently low that a grid is not used.
A slot-beam system for digital mammography was proposed by Nelson et al. (1987). A prototype slot-beam system was developed (Maidment and Yaffe, 1990; Nishikawa et al., 1987; Tesic et al, 1999; Yaffe, 1993; Yaffe et al., 1996) and designed to operate with an acquisition time that is acceptable for clinical imaging. A clinical system of this type introduced by Fischer Imaging, Inc. (Denver, Colorado) has received FDA approval. After transmission through the breast, x rays are absorbed by a cesium-iodide phosphor, and the emitted light is conveyed via fiber-optic couplers to several CCD arrays whose electrical signals are then digitized. This design can provide 50 pm sampling (25 pm for a partial image) referred to the midplane of the breast. The restricted angular acceptance of the optical fibers causes each fiber to collimate the light incident from the screen, thereby increasing the effective resolution. In addition, the high-optical coupling efficiency attainable with fiber optics minimizes signal losses, thereby facilitating an x-ray quantum limited system.
The image is acquired by scanning the fan x-ray beam and the slot detector across the breast in a direction parallel to the short dimension of the detector. To allow a smooth mechanical motion, the images can be acquired using a time-delay integration technique (Holdsworth et al., 1989). As the detector is moved across the breast at constant speed, the charge collected in each element of the CCD is shifted down its column at the same speed as the scan, but in the opposite direction, resulting in integration of the signal corresponding to a given image pixel. When the charge packet reaches the last element in the CCD, the charge signals in the columns are read out. Depending on the slot width, a scanning system can acquire a mammogram in 3 to 6 s.
Selenium has very high electrical resistivity in the dark, so that if a plate of selenium is uniformly charged, the charge will remain in place on the surface. When exposed to x rays, the plate will discharge; the degree of discharge being proportional to the amount of radiation striking the plate. For digital mammography, the selenium can be deposited on an array of electrodes where each element contains a collector electrode and thin film transistor or diode switch for readout, in a manner similar to that of the amorphous silicon system described above. A system of this design produced by Lorad Corporation has received FDA marketing approval and is shown schematically as Type 5 in Figure 3.23. Two companies, Instrumentarium Imaging and Siemens Medical Solutions are producing units with amorphous selenium. Note that in both Type 3 and Type 5 detectors, the photoiodide or collection electrode are co-planar with the transistor switches, although for clarity they are drawn at separate levels in the cross-sectional schematic.
The electronic (soft-copy) display of images on digital mammog-raphy systems is limited by the spatial resolution of currently affordable displays (Feig and Yaffe, 2005; Yaffe, 1999). A basic requirement for general use is the ability to portray the entire breast with sufficiently fine detail so that tiny structures (e.g., microcalcifications) indicative of malignancy are readily visible. Furthermore, since routine mammographic interpretation involves viewing four images of a current examination compared with four images from a prior examination, digital work stations must permit simultaneous display of these eight images, using either eight networked monitors or, a lesser number of monitors providing sufficiently fine detail that two or more whole-breast mammograms are displayed per monitor (Huang and Lou, 1999; Lou et al., 1994). Because soft-copy display technology is currently not able to meet these requirements for systems that provide pixels smaller than 100 pm, the development of innovative methods for rapid image navigation and manipulation is a high priority.
The technique (operating potential, filtration, etc.) for screen-film mammography has been established, largely by trial and error, over several decades of the practice of mammography. For digital systems where contrast can be freely manipulated, the optimal spectra may be different than for film. In a digital imaging system, the operating potential and the amount of radiation used to form an image should be defined strictly by signal-to-noise considerations rather than, by contrast or film "blackening." Increased operating potential, compared to screen-film technique, improves efficiency and output of the x-ray tube resulting in images with a higher SNR, while allowing reduced patient dose and scan time. It is important to ensure that digital mammography techniques are appropriately optimized for those imaging tasks being considered so as to take advantage of the possible performance gains that digital mammography may provide.
The evaluation of digital mammography is still underway. Its performance can be properly evaluated only in careful studies that compare its sensitivity and specificity to that of high-quality screen-film mammography. During those studies, the quality of both the conventional and digital imaging must be carefully monitored and controlled (Lewin et al., 2001; 2002). A large clinical study, Digital Mammography Imaging Screening Trial, is currently being carried out in the United States and Canada. In the trial, 49,500 women at 34 sites will receive both screen-film and digital mammograms and the accuracy of the two methods will be compared (Pisano et al., 2000). Screening trials comparing conventional and digital mammography are further discussed in Section 8.
The principal theoretical advantage of digital mammography comes from decoupling image display from image acquisition. This permits the digital image to be acquired, stored electronically, and then manipulated, analyzed and displayed as needed. It is anticipated that digital mammography will provide improved visualization of the structures within the dense breast, thereby increasing the value of mammography in those women. Even if the sensitivity and specificity are only equal to, but not better than screen-film mammography, digital mammography is still likely to play an important role in the detection, diagnosis and management of breast cancer. This statement is based on the potential value of applications that will be greatly facilitated through the availability of mammograms in digital form. These include increased throughput, computer-aided detection/diagnosis (CAD), telemammogra-phy, automated QC, image processing, more efficient archiving and retrieval, and the availability of dual energy, stereoscopic and tomographic methods.
These CAD applications can be used by radiologists as second interpretation devices to avoid overlooking identifiable mammo-graphic abnormalities (Chan et al., 1990; Giger, 1999). This approach will be much less expensive than a second reading done by another radiologist, but only if the false-positive detection rate of computer-identified findings decreases substantially from current levels. Ultimately, highly sensitive lesion detection applications might be used for the first-pass interpretation of digital mammography screening examinations, forwarding only those cases with suspect findings on to a radiologist for definitive interpretation and, if necessary, for further imaging evaluation.
Numerous clinical studies have shown that the detection sensitivity of CAD is higher for calcifications (83 to 100 percent) than for masses (34 to 95 percent) (Feig and Yaffe, 2005). Masses that do not contain calcifications or spiculation that is not prominent are less likely to be identified by CAD. Cancers in which the mass appears subtle to radiologists or also looks like an architectural distortion or asymmetrical density rather than a mass are also less likely to be flagged by CAD (Vyborny, 2000). Several studies have found that CAD may increase radiologist's cancer detection rates by as much as 20 percent (Destounis et al., 2004; Freer and Ulissey, 2001). However, because of the significance of false-negative findings (missed breast cancers) and because it is unlikely that software vendors will assume any medicolegal responsibility for their CAD programs, it is equally unlikely that this software will be used for first-pass interpretation.
Computer-aided interpretation programs also are being developed to further characterize already detected lesions to aid the radiologist in determining whether subsequent management should involve biopsy or less invasive procedures (Ackerman and Gose, 1972). Again, efforts have been directed principally at the analysis of clustered microcalcifications (Chan et al., 1998; Fox et al., 1980; Goumot et al., 1989; Jiang et al., 1999; Magnin et al., 1989; Wee et al., 1975). Applications operate by quantifying the digital data within suspect lesions that already have been flagged, either by radiologists or by CAD programs. Formulas, then, are used to analyze a wide variety of lesion characteristics for calcifications. These can include not only the standard parameters assessed by radiologists (particle size, number, distribution, density and shape), but also several more complex measures of calcification irregularity (e.g., compactness, eccentricity, coefficient of convexity, elongation) (Goumot et al., 1989; Magnin etal., 1989). Finally, numeric scores calculated for these parameters are weighted by predetermined algorithms and combined to produce a likelihood of malignancy index, upon which management decisions can be based. Currently, the most successful of the calcification characterization programs perform at diagnostic accuracies that approximate, and occasionally even exceed those of expert mam-mographers (Feig and Yaffe, 2005; Giger, 1999; Giger et al., 2000). For breast masses and other types of suspect lesions, today's CAD programs are less fully developed and also somewhat less successful (Huo et al., 1998; Kegelmeyer et al., 1994; Patrick et al., 1991; Yin et al., 1993).
Computer analysis of digitized mammograms can also be used to extract other valuable information such as the future risk of developing breast cancer. Boyd et al. (1998), have demonstrated a strong correlation between breast density and breast cancer risk by simple thresholding of digitized mammograms.
22.214.171.124 Computer-Aided Instruction. Rapid and inexpensive computer-based storage of digital mammography examinations facilitates the creation and utilization of computer-aided instruction packages, since preselected sets of images can be readily catalogued and retrieved for display. The simplest application is the digital counterpart to the conventional film mammography learning file. This consists of an organized library of interesting case material (digitized mammograms), supplemented by hard-copy text descriptions of mammographic findings, suggested interpretations, pathologic correlations, additional discussions, and literature reference material for each case or group of cases. Many mammography cases can be stored on a single CD-ROM. In more sophisticated systems, the text material itself is stored electronically so that cases can be viewed either in random sequence (as unknown cases) or, in sequences ordered either by diagnosis or by specific mammographic finding.
Instructional programs also are being developed to provide the user with response-driven self-instruction courses in which incorrect answers trigger the display of remedial material and additional questions before subsequent cases can be viewed (Cao et al., 1997). These systems can track the progress of individual users, compiling grades and documenting that proficiency has been achieved.
The ultimate instructional package will interface directly with the day-to-day interpretation of digital mammograms. Such a program would be activated either by request of the radiologist or, whenever computer-recorded interpretations indicate specific mammographic findings. In either circumstance, description of particular mammographic findings would call up related image and text materials from expert learning databases to aid in the analysis of the case under consideration (Swett and Miller, 1987; Swett et al., 1989). Thus, the radiologist could simultaneously view pathology-proven cases in which mammograms display similar, if not identical radiographic findings. Embedded text also could suggest predetermined strategies for further evaluation and interpretation of the mammographic findings.
The principal deficiencies of current digital mammography equipment involve limitation in the capabilities of existing soft-copy display systems to rapidly display images of the present and/or previous examinations with the full acquired spatial resolution. There are also practical challenges in displaying the information in a manner that allows interpretation to be as rapid and efficient as is now possible with current view-box presentation. Cost of the display systems is also a concern. Digital mammograms can also be read from laser-printed films which provide adequate dynamic format and spatial resolution, but do not have adequate dynamic range to depict all of the information in the digital image in a single presentation unless appropriate image processing is performed prior to printing. Teleradiology applications will benefit from improved software techniques to compress and store digital data, as well as from development of more efficient protocols to accelerate image transmission. Computer-aided diagnosis applications also will continue to increase in accuracy as existing algorithms are refined and new ones are developed, driven at least in part by the use of neural networks and other forms of machine intelligence (Wu et al., 1993).
While mammography provides high sensitivity to the detection of breast cancer, the only definitive test for breast cancer is biopsy. Breast cancers missed by mammography can be minimized only if a diagnostic threshold is used that incurs a reasonable number of false positives. As a result, two-thirds to four-fifths of surgical breast biopsies yield negative results. Over the last decade, stereo-tactically-guided and ultrasound-guided core needle biopsies have become the standard of care for tissue sampling of suspicious breast lesions. This has been due to pioneering work that has demonstrated that large-core sampling can replace open excisional breast biopsy in most patients. Development and clinical implementation of prone stereotactic biopsy systems with digital image receptors have made the procedure faster, more reliable, more comfortable, and less traumatic for the patient.
Stereotactic breast biopsy uses the principle of parallax: two planar radiographic views acquired at different x-ray source positions are used to determine the location of radiographically visible objects in three dimensions. Dedicated prone biopsy systems place the patient in the prone position with the breast dependent through a hole in the table. The x-ray tube, compression device, image receptor, and biopsy device are mounted under the table, which can be raised to make more working space for the physician conducting the procedure. With the breast compressed, stereotactic views are acquired and targeted lesions are marked in both views to direct needle placement. A precise mounting system (called the punction device or staging unit) is used to hold the core biopsy device and direct sampling to the desired location within the breast.
The development that brought renewed interest to stereotactic localization was the acquisition of core biopsies using prone positioning directed by stereo x-ray images using larger (14 gauge) cutting needles. The cutting needle biopsy consists of a double cannula needle system operated by a biopsy gun. The gun-needle system first deploys the inner needle containing a sampling notch. As the needle is rapidly pushed forward, the beveled tip deflects the needle, opening the sampling notch to tissue. An outer cylindrical cannula then deploys, slicing off a small segment of tissue that is retained in the sampling notch. The deployment (firing) of the two-component needle system is an integral part of tissue sampling in the cutting needle approach. A mounted biopsy device is used to hold and deploy the 14-gauge cutting needle. After each sample is acquired, the cutting needle must be removed from the breast, the biopsy device removed from the holder, and the needle removed from the biopsy gun to remove the tissue sample. Since multiple tissue samples are required, this process must be repeated. At least five samples are recommended for soft-tissue lesions, and at least ten samples for targeted calcifications.
Core needle biopsies acquired with this system require only a small amount of local anesthetic and a small incision at the point of needle entry into the skin. Core needle biopsies take about one-half hour to perform, are one-fourth to one-half the cost of surgical excisional biopsies, involve minimal risk, and produce no residual scarring of breast tissue. Placing the patient in the prone position rather than upright during biopsy, as with stereotactic devices added on to standard mammography units, minimizes patient motion during localization, eliminates vasovagal reactions (fainting), and provides more working space for the radiologist during the biopsy procedure.
Stereotactic breast biopsy has been improved even further by the development of the vacuum-assisted core biopsy system. The first vacuum-assisted core biopsy system, the Mammotome system, was developed by Burbank and Parker (Burbank, 1997; Burbank et al., 1996). The Mammotome system uses a double cannula needle: the outer needle is hollow, with a sampling notch near the end and a vacuum system that pulls tissue into the sampling notch, when open. The inner needle is a hollow cylindrical cannula that can be pulled back to expose the sampling notch, then rotated and advanced to cut off a cylinder of tissue drawn into the sampling notch by the vacuum. A second vacuum line is used to retain the cylinder of tissue at the end of the inner cannula as it is pulled through the outer needle. After the sample is captured, but before it is removed, the sampling notch can be rotated to a slightly different angle to prepare for removal of additional tissue samples. This design permits removal of the tissue sample without having to remove the outer needle from the breast.
The Mammotome system accommodates either 14-gauge or larger 11-gauge needles. The 14-gauge Mammotome system acquires tissue samples that are two to five times larger than 14-gauge cutting needles; the 11-gauge Mammotome system acquires tissue samples that are 5 to 14 times larger than 14-gauge cutting needles. Another advantage of the vacuum system is the removal of blood and fluid, so that samples consistently contain solid tissues. Because of the marked improvement over the cutting needle approach, >90 percent of core biopsies are now performed using the suction biopsy system.
Other vacuum-assisted biopsy systems have recently been introduced. Like the Mammotome system, tissues can be retrieved without having to remove the entire needle device from the breast. Also like the Mammotome system, larger tissue samples are routinely retrieved than with a 14-gauge cutting needle and tissue samples are less likely to consist entirely of blood or fluid.
Prone stereotactic systems now employ digital image receptors based on CCD arrays. A CCD array is a small panel of lightsensitive diodes. CCD arrays are used as the video pickup in modern video cameras. The arrays used are typically 1 x 1 inch arrays with 1,024 x 1,024 matrix elements. X rays are intercepted by a fluorescent screen that converts each absorbed x ray to thousands of visible light photons. A small fraction of the emitted visible light photons is absorbed by the CCD array. Different systems use different methods of directing the emitted visible light photons toward the CCD array. The Fischer system uses a fiber-optic taper with 2:1 demagnification to direct photons emitted from the exit surface of a gadolinium-oxysulfide screen or more recently a cesium-iodide crystal screen to the CCD array. The Lorad system uses a set of mirrors and lenses with approximately 2:1 demagnification to direct photons emitted from the entrance surface of a gadolin-ium-oxysulfide screen to the CCD array.
For small lesions, there is a risk that core biopsy will remove all radiographically- or ultrasonically-visible signs of the lesion. To guide surgical removal of the remaining lesion and margins in cases where the removed tissue is analyzed to be malignant by pathology, radiographically or ultrasonically visible markers can be inserted through a larger vacuum-assisted biopsy needle at the time of tissue removal. A number of manufacturers have developed small metallic marking clips for radiographic marking, or gel-markers for ultrasound marking, of biopsy sites.
Was this article helpful?