| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
1 From the Duke Advanced Imaging Laboratories, Departments of Radiology, Physics, and Biomedical Engineering, Duke University, DUMC Box 3302, Durham, NC 27710. From the AAPM/RSNA Physics Tutorial at the 2003 RSNA Scientific Assembly. Received October 4, 2004; revision requested November 3 and received December 6; accepted December 13. The author has no financial relationships to disclose. Address correspondence to the author (e-mail: samei{at}duke.edu).
 |
Abstract |
|---|
 Top Abstract IntroductionDisplay TechnologiesPsychophysical FactorsTesting Mammographic DisplaysConclusionsReferences |
|---|
© RSNA, 2005
Abbreviations: CRT = cathode-ray tube, LCD = liquid crystal display
|
Introduction |
|---|
|
TopAbstractIntroductionDisplay TechnologiesPsychophysical FactorsTesting Mammographic DisplaysConclusionsReferences |
|---|
To justify the transition from the less costly analog mammography to the more costly digital alternative, many advantages are often cited. Among these are higher image quality, higher efficiency, potential for image manipulation, and remote display of images. While it is possible to display digital images in a hard-copy format on transilluminated film, as was often done in early implementations of digital technology, these advantages cannot be fully realized without the soft-copy display of digital mammograms, making electronic display a key and inseparable component of digital mammography. As such, the quality of the display device can have a direct impact on the effectiveness of the mammography operation. A faulty or improperly set up display device can compromise the overall quality of the diagnostic procedure (1–5). This factor is particularly important considering the current suboptimal performance of electronic display media in comparison to conventional transilluminated films (6).
The purpose of this article is, first, to briefly outline the two general technologies that are currently used for digital mammographic displays, namely cathode-ray tubes (CRTs) and liquid crystal displays (LCDs). Second, some of the psychophysical issues pertaining to effective soft-copy viewing of mammograms will be discussed. These include factors related to human visual acuity as well as presentation considerations from an ergonomic and cognitive standpoint. Finally, the article ends with a brief reference to testing methodologies for digital mammographic displays.
|
Display Technologies |
|---|
|
TopAbstractIntroductionDisplay TechnologiesPsychophysical FactorsTesting Mammographic DisplaysConclusionsReferences |
|---|
Cathode-Ray Tubes
Based on a mature technology nearly a century old, CRTs produce soft-copy display of images by continuous "painting" (ie, raster scanning) of a phosphor screen with a scanning electron beam (Fig 1). Initially, an electronic beam is generated by an electronic gun located at one end of a vacuum tube. The beam is accelerated and focused on a phosphor screen located at the other end of the vacuum tube via a series of electrostatic lenses. A magnetic deflection yoke is used to alter the path of the beam into a periodic raster scanning pattern, scanning the phosphor screen from left to right and from top to bottom at a rate of 60–80 frames per second. As the beam is scanned across the screen, its intensity is modulated in correspondence with the beam’s location on the phosphor screen and the image being displayed. The phosphor screen emits light in proportion to the intensity of the electronic beam impinging on it, and the emitted light forms the image on the faceplate of the CRT. In this manner, the image is continuously "painted" on the CRT phosphor screen in a spatially continuous (analog) fashion.
|
Liquid Crystal Displays
LCDs use a technology markedly different from CRTs. Taking advantage of the light polarization properties of liquid crystals in conjunction with thin-film transistors (TFTs), LCDs produce soft-copy display of images by modulating light transmission through the minute "windows" (ie, liquid crystal cells) of a flat-panel display. The display is made of a series of thin components coupled to one another (Fig 2). On the back side, there is a uniform fluorescent light source. The light is polarized by a back polarizing filter and transmitted through a glass substrate, on which a layer of TFTs is deposited. Each individual transistor can apply a specified voltage, controlled by horizontal and vertical gate lines embedded in the display structure, to the liquid crystal cell adjacent to the TFT. Depending on the magnitude of the applied voltage, the polarization orientation of the liquid crystal changes, affecting the transmission of already polarized backlight through the cell. The light transmission and therefore also the luminance of each individual cell is unique. In this manner, the image is formed on the display surface in a spatially discrete (ie, digital) fashion. Additional elements of the display structure include spacers (to keep the thickness of the liquid crystal cell uniform), an output polarizing filter (for better light transmission control), and a faceplate.
|
|
Psychophysical Factors |
|---|
|
TopAbstractIntroductionDisplay TechnologiesPsychophysical FactorsTesting Mammographic DisplaysConclusionsReferences |
|---|
Display Luminance and Contrast Factors
Luminance, or perceived brightness, is generally understood to be an important factor in display quality. Higher-brightness images cause a larger number of visible photons impinging on the retina, reducing noise in the human visual system and thus improving the contrast threshold. Therefore, mammographic displays should have a minimum level of maximum brightness in the range of a few hundred candelas per square meter. However, a higher-luminance display device does not necessarily suggest a higher-quality display. As the human visual system is able to adapt to different brightness conditions, the main quality of concern is not the absolute magnitude of the luminance but rather the change in luminance characterized in terms of contrast.
The maximum contrast achievable by a display device is often characterized in terms of contrast ratio (or luminance ratio [LR]) as the ratio of the maximum to minimum luminance. Two display devices with different absolute luminance levels offer comparable qualities if their LRs are similar. To match the current standard of practice, a mammographic display should have a minimum luminance ratio comparable to typical brightness ranges seen on radiologic images. If it is assumed that film density interpretable without the aid of a high-brightness illuminator ranges between 0.1 and 2.5, that minimum LR would be 250. Display devices with an LR less than this limit will offer perceived contrast lower than transilluminated film, while those with significantly higher LR values can render the dark regions of the image invisible due to the limited adaptation state of the human visual system and the contrast reduction processes of veiling glare and reflection (discussed later).
As the preceding discussion implies, the basic psychophysical quantity luminance (L) is best characterized in terms of contrast (
L/L or dL/L). Contrast is related to the difference in the luminance of an image feature of interest in comparison to the background luminance. The contrast response of the human visual system is often characterized in terms of the contrast threshold, defined as the minimum change in luminance needed to achieve a just noticeable difference in the image (ie, dL/L for a JND). In situations in which the human visual system is presented with low-contrast features imbedded in an otherwise uniform background and the eye is adapted to the average luminance of the scene, the human visual system exhibits a nonlinear contrast threshold that is generally reduced with increased luminance (Fig 3a).
|
|
In radiographic imaging, including mammography, the logarithm of x-ray transmission is generally proportional to tissue thickness/density. Thus, if the system has a logarithmic response to exposure, and the logarithmic image values are mapped to luminance values such that the logarithm of luminance is linearly proportional to the image value, the luminance values representing the image will be linearly related to tissue thickness/density. This is essentially the case for transilluminated mammography films viewed on light boxes where the luminance range falls within the linear-log portion of the GSDF. However, current soft-copy displays have lower luminance values, and thus, given the nonlinear response of the human visual system at lower luminance values, the contrast at darker regions of the image must be boosted to maintain the same thickness dependency. This contrast enhancement at low luminance values can be achieved by GSDF calibration.
While displaying a medical image with a luminance response according to the GSDF can provide a coarse similarity in image appearance across various display devices (as is the premise of the so-called perceptual linearization), the images can still vary noticeably from display to display. This variation can be attributed in part to the adaptation processes of the human visual system: The Barten model is based on experimental data in which the eye is adapted to the luminance value of a uniform background, the state of so-called variable adaptation. In actual medical images, on the other hand, each image presents a wide range of luminance values and the eye is adapted to the global average luminance of the scene, the state of so-called fixed adaptation. Furthermore, this fixed adaptation can change depending on the luminance in the local area surrounding each point of eye fixation. Figure 4 shows the contrast response of the human visual system in the state of fixed adaptation for displays with different luminance ratios, neglecting the secondary local adaptation effects. In all examples shown, it is assumed that the eye is adapted to the average log luminance in the scene. It is evident that even if a display system is calibrated to the GSDF standard and even if the luminance ratios are similar, the difference in the contrast sensitivity of the human visual system based on the mean luminance would render images different.
|
Display veiling glare similarly reduces display contrast. Display glare refers to diffuse spreading of the luminous signal caused by light scattering in the display’s faceplate, light leakage (particular to CRTs), electron backscattering (prominent in color CRTs), or electronic cross-talk (in LCDs). The resulting effect is a low-frequency scene-dependent degradation of image contrast, particularly of features in dark regions surrounded by bright scenes. It is best to use display devices with minimal glare for mammographic applications.
Two other psychophysical factors impact the luminance features of soft-copy displayed mammograms: luminance nonuniformity and angular dependency. When one considers that the display area of most medical display devices extends to dimensions of 30 x 40 cm, it is not trivial to maintain luminance uniformity across such a large area. The main causes of variation are changes in the electronic beam profile and face-plate thickness in CRTs and backlight and liquid crystal thickness nonuniformity in LCDs. While for a typical medical-grade display device, the luminance across the faceplate can vary by as much as 10%–30%, luminance nonuniformity is rarely considered a major quality issue. This is because of the fact that the human visual system has very low sensitivity to broad low-frequency variations, as shown in Figure 5. If the luminance fluctuations range between one to one-half of the display size (0.003–0.007 cycles per millimeter for a 30-cm display area at a 30-cm viewing distance), the contrast threshold of the human visual system at the corresponding spatial frequencies is within the 19%–43% range for viewing at a 30-cm distance and the 9%–19% range at a 60-cm distance, meaning that nonuniformities of such scale with contrasts less than the indicated values would not be readily discernible.
|
Resolution Factors
The luminance response of the human visual system described earlier pertains to the response at only one spatial frequency. The human eye responds differently to different spatial frequencies. Figure 5 illustrates the spatial frequency response of the human visual system in terms of the modulation threshold at close inspection (30-cm) and normal viewing (60-cm) distances. It is evident that increasing the mean luminance value lowers/ improves the contrast threshold. In the 30–60-cm viewing distance range, the lowest threshold (optimum sensitivity) occurs in the 0.4–1.2 cycles per millimeter frequency range, corresponding to image features 0.4–1.3 mm in size. Below and above this range, the threshold increases.
One of the questions often asked about mammographic displays is the display pixel size desirable to achieve high-fidelity display of the images. Clearly, smaller pixel sizes will offer more detailed representation of the image. However, the reduced sensitivity of the human visual system at high spatial frequencies will render the highest levels of detail invisible. Ideally, the level of detail presented should match the detection threshold of the human visual system. By using a conservative 5% contrast as the detection threshold of the human visual system, it is possible to determine the limiting resolution and pixel sizes that correspond to this threshold. In Figure 5, the spatial frequencies associated with a 5% contrast threshold would be in the range of 2.3–2.6 cycles per millimeter at a 60-cm viewing distance and 4.8–5.5 cycles per millimeter at a 30-cm distance. These frequency ranges correspond to 0.19–0.22-mm and 0.09–0.10-mm pixel sizes, respectively. If one assumes a typical 30 x 40-cm display area, these pixel sizes relate to display matrix sizes of 2.5–3.2 megapixels and 11–15 megapixels, respectively. The current commercial display technology offerings readily cover the first range but fall short of approaching the second one. The recent requirement of 5 megapixels as a minimum matrix size for digital mammographic displays by the U.S. Food and Drug Administration (FDA), while satisfying the first condition, falls short of satisfying the latter one.
Apart from the nominal size of the display pixels, the spatial extent of the pixels also has a psychophysical influence on presentation quality. This is more of an issue in CRT displays, where display pixels have a pseudo-Gaussian profile extending beyond the nominal pixel size. It is possible to characterize the resolution of a display in terms of the modulation transfer function (MTF), described in other publications (11). While the relationship between display resolution and diagnostic accuracy has not been fully substantiated, it is generally understood that the higher the MTF, the better the sharpness and image quality of the displayed mammogram.
Noise Factors
The third psychophysical element of image quality is noise. Display noise refers to luminance fluctuations (either spatial or temporal) that are added to the image by the display system. Temporal noise is often of less concern than spatial noise in mammography as the human visual system integrates the visual signals in the temporal domain, reducing the perception of fluctuations in static mammographic images. The only exception to this statement is flicker, which can cause fatigue and discomfort. Spatial noise has been well characterized as a source of error in radiographic imaging. The classic Rose model (12) describes how the detectability of low-contrast features against a noisy background is directly related to the level of noise in the image. Specifically, the threshold contrast-diameter product increases linearly with the magnitude of relative noise.
By using the Rose model, it is possible to establish the noise threshold below which the noise due to display can be considered negligible. Ideally, a display should not add luminance fluctuation to an image beyond inherent fluctuations already present in the image. In digital mammography, the average noise is in the neighborhood of 2%–2.5% (assuming a 46,052-mm–2mR–1 ideal SNR2, 10-mR average detector exposure post breast, 0.7 detective quantum efficiency at zero frequency, and 0.07–0.1-mm pixel size). Aiming for a display device that will not add more than a third of the noise of a typical mammogram limits the display relative noise to 0.6%–0.8%. By using the Rose model with the threshold SNR of 3, this noise level corresponds to a contrast threshold of 0.02–0.03 mm in terms of the contrast-diameter product.
In current display technologies, two types of noise may be recognized: luminance noise and structured noise. Luminance noise refers to random spatial fluctuations in the pixel luminance values when a uniform pattern is displayed. In CRTs, luminance noise is dominated by phosphor granularity. The two common types of phosphors used in medical CRTs, p45 and p104, have markedly different noise characteristics. The single-component p45 phosphor, while less efficient in terms of light yield, offers notably lower noise levels than multicomponent p104 phosphor. In LCDs, luminance noise is generally low and is dominated by electronic noise as well as pixel-to-pixel variations due to nonuniformity of pixel gain across the display (13). Recent studies have suggested that such variations can be reduced by a gain calibration procedure, not much unlike the gain calibration used for digital radiographic detectors.
Structured noise refers to spatially fixed, recognizable structures within the displayed image. The main source of structured noise in monochrome CRTs is the raster lines. An artifact of the electron beam scanning mechanism of CRTs, these horizontal lines create a fixed pattern in the displayed area (Fig 6a). In LCDs, the main source of structured noise is the pixel structure. Owing to the construction of the LCDs, a large fraction of the display area does not transmit the backlight. The opaque areas create a periodic pattern in the image. The pattern is much more pronounced than raster lines in CRTs since it is created by very large dark-to-bright modulations (Fig 6b). When the magnitude of noise alone is considered, LCD-displayed images are markedly noisier, mostly due to structured noise. However, it is uncertain if structured noise influences detectability the same way the random noise does. For example, there are indications that human observers are able to discount known structures from an imaging scene through the so-called pre-whitening process. More research is needed to draw definite conclusions about the impact of LCD structured noise on display quality.
|
|
In a typical mammography case interpretation, up to eight or even 16 mammograms may need to be reviewed, corresponding to current and prior craniocaudal (CC) or mediolateral oblique (MLO) mammographic views of left and right breasts. Initially, a global view of at least four of these images is used. In terms of presentation ergonomics, there are currently no universally accepted formats for displaying these many mammograms. Some viewers prefer to see the images of the left breast on the left and those of the right breast on the right, while others prefer otherwise. The four images may need to be configured four different ways, depending on the MLO/CC laterality preferences. Additional prior views further complicate formatting. Given this large potential variability, a mammographic workstation should be highly configurable, enabling the user to set up the default presentation according to his or her preferences.
In terms of the number of required monitors per workstation, while many images may need to be viewed in an interpretation session, multi-image comparisons rely mostly on gross differences in the images and usually no more than two images may need to be compared at high resolution at a given time. Furthermore, there are advantages to comparing images temporally by toggling between images as opposed to spatially by viewing images side by side. This is due to the fact that the temporal channels of the human visual system are more sensitive than the spatial ones. Thus, two monitors per workstation is usually optimum. Beyond four monitors, the efficacy of the interpretation will become suboptimal.
In terms of functionalities, the user interface should enable full visualization of image data by magnification and zoom functionalities for full "probing" of the spatial domain; windowing and leveling for full probing of the intensity domain; image processing; and perhaps advanced processing such as computer-aided detection (CAD). However, the workstation should provide these functionalities without negatively impacting observer performance. The interface should be intuitive, requiring minimal thought to navigate, thus enabling more attention to be paid to the images themselves. The interface should ideally imitate current practice and be nonintrusive, with only small low-contrast "real estate" devoted to functional buttons. It should also require minimum user interaction by having multitask shortcut buttons and few "dragging" operations. It may also be useful if the interface can offer user-defined structured "directed viewing," in which the mammographer is guided through a predetermined optimized set of viewing sequences for interpreting the images.
|
Testing Mammographic Displays |
|---|
|
TopAbstractIntroductionDisplay TechnologiesPsychophysical FactorsTesting Mammographic DisplaysConclusionsReferences |
|---|
While acceptance testing establishes the initial performance of a display device, the performance can change over time. In CRTs, notable change can occur in the luminance, geometric characteristics, and resolution characteristics of the device. In LCDs, luminance can similarly change over time. The effective performance of both types of displays can also be obviously influenced by changes in the ambient lighting conditions. Acceptance testing provides baseline data to be used for monitoring the quality of the display device to ensure that a high-quality mammography operation is maintained over time. Monitoring display quality should take place in the context of a quality control (QC) program, which should include annual inspections, quarterly or monthly evaluations, and daily tests. Within this spectrum of testing, the tests range from objective quantitative tests at acceptance and annual testing to a fast visual test for daily inspection.
In any digital mammography operation, two classes of display devices may be identified: primary and secondary. Secondary devices are those used by technologists to review and possibly process the images after acquisition, as well as those used to view images for which the official interpretation has already been rendered. The primary devices are those used for primary interpretations of mammograms. While the quality requirements for the two classes of devices vary, both types of devices should adhere to certain minimum requirements.
The quality assurance of mammographic displays in the United States is regulated under the Mammography Quality Standards Act (MQSA) (21 CFR 900) through each vendor’s quality control program, overseen by the Food and Drug Administration (FDA) and the American College of Radiology (ACR). There are a number of standards for assessing the performance of electronic display devices, most notably those formulated by the Society of Motion Picture and Television Engineers (SMPTE) (14), Digital Imaging and Communications in Medicine (DICOM) (8), the German Institute for Standardization (Deutsches Institut für Normung [DIN]) (15), the Video Electronics Standards Association (VESA) (16), and the American Association of Physicists in Medicine (AAPM) Task Group 18 (TG18) (17). The vendors’ quality control programs are generally based on these standards, even though they are not currently applied uniformly.
The most recent display quality guidelines, AAPM TG18, are particularly applicable to acceptance testing and quality control of mammographic displays. The TG18 tests include methods for assessing the luminance and contrast performance of mammographic displays, taking into consideration the psychophysical factors discussed earlier. The guidelines also suggest minimum expected performance values. Similarly, further tests and criteria include display resolution and noise as well as other display characteristics such as angular response, reflection, glare, distortion, color tint, and artifacts. The guidelines also recommend a strategy to ascertain the maximum allowable illumination in the viewing area by using the reflection and luminance characteristics of the display. Most of the recommended tests use newly developed test patterns (Fig 7). There have been a number of recent articles summarizing the TG18 methodology (18). Interested readers are advised to consult these publications or the TG18 document itself (17) for detailed description of the testing methods.
|
|
Conclusions |
|---|
|
TopAbstractIntroductionDisplay TechnologiesPsychophysical FactorsTesting Mammographic DisplaysConclusionsReferences |
|---|
|
Footnotes |
|---|
|
References |
|---|
|
TopAbstractIntroductionDisplay TechnologiesPsychophysical FactorsTesting Mammographic DisplaysConclusionsReferences |
|---|
This article has been cited by other articles:
|
|
 |
|
  D S Brettle Display considerations for hospital-wide viewing of soft copy images Br. J. Radiol., July 1, 2007; 80(955): 503 - 507. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 B. F. Branstetter IV Basics of Imaging Informatics: Part 1 Radiology, June 1, 2007; 243(3): 656 - 667. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOGRAPHICS | RADIOLOGY | RSNA JOURNALS ONLINE |
