(Radiographics. 1999;19:1313-1318.)
© RSNA, 1999
A Perceptually Tempered Display for Digital Mammograms1
Harold L. Kundel, MD ,
Susan P. Weinstein, MD ,
Emily F. Conant, MD ,
Lawrence C. Toto, BS and
Calvin F. Nodine, PhD
1 From the Pendergrass Diagnostic Research Laboratory, Department of Radiology, University of Pennsylvania Medical Center, 308 Stemmler Hall, 3600 Hamilton Walk, Philadelphia, PA 19104. Recipient of a Certificate of Merit award for an infoRAD exhibit at the 1998 RSNA scientific assembly. Received April 9, 1999; revision requested May 12 and received June 10; accepted June 21. Supported by grants DAMD17-96-1-6153 and DAMD17-97-1-7130 from the U.S. Army Medical Research and Materiel Command. Address reprint requests to H.L.K.
 |
Abstract
|
|---|
The cathode ray tube of a workstation for use with digital mammograms was calibrated with a photometer to produce an input-output characteristic curve similar to the perceptually linear curve defined by a current display standard. Then, a test pattern consisting of bars of increasing intensity containing disks of decreasing contrast was used by an observer to estimate the minimal detectable contrast (MDC) at different levels of display luminance. The MDC was modeled by a parabola. The shape of the parabola was determined by the observer's perceptual responses, and the range was determined by the maximum and minimum pixel values of the breast parenchyma. As each mammogram was displayed, the contour of the breast was automatically found and pixels within the breast image were sampled to determine the pixel values that were used to compute the maximum and minimum pixel values. The parabola was integrated to determine the look-up table for the initial MDC-tempered display of the mammogram. Preliminary observer performance tests showed no significant differences in the accuracy and speed of three radiologists who read a set of mammograms when the MDC-tempered display was compared with the perceptually linear display.
Index Terms: Breast, 00.99 • Images, display • Radiography, digital
|
INTRODUCTION
|
|---|
Given the present state of the art, a static cathode ray tube (CRT) display can simulate but not duplicate the image quality of a film mammogram displayed on a light box. The film is displayed at higher luminance and has greater spatial resolution and a wider gray-scale range (1). However, the film captures and displays the image by using a fixed set of predetermined display parameters. The CRT display can be adjusted to explore the full range of contrast and resolution available in the digital image by using the window level to change the gray-scale range and zoom-rove functions to change spatial resolution. The appearance of the gray scale within the image can also be changed by modifying the input-output transfer characteristic of the CRT by using look-up tables. The overall appearance of the image can also be changed in more fundamental ways by the application of image processing such as edge enhancement. In this article, we consider only the effects of modifying the input-output transfer characteristic. To have identical images look alike when displayed on different CRTs, a display standard called perceptual linearization has been proposed (2,3). When the standard is used, equal changes in the pixel gray-scale value produce equal changes in the just noticeable difference (JND) of luminance in the image.
A display standard provides an equivalent starting place for each image but may not provide the best distribution of gray levels for a particular image in a particular reading environment. For example, the image may be too dark or too light, just as an image on film may be under- or overexposed. The ability of the human eye to see the intensity difference between two areas in an image (contrast sensitivity) depends on the average intensity of the light reaching the eye (4). The average intensity of the light reaching the eye is termed the adapting luminance. When the adapting luminance is very different from the average luminance of the area of interest in the image, the ability to see contrast is decreased. This is the reason why masking the bright areas on a film illuminator improves the appearance of images, particularly dark ones. Most of the light that affects contrast sensitivity comes from the displayed image, but some comes from room illumination including that which is reflected from the CRT surface. Once the room illumination has been minimized, the contrast sensitivity of the eye can be maximized by adjusting the gray scale to smooth out extreme variations in brightness within the image (5,6).
Using a model proposed by Mokrane (7), Liu and Nodine (8) developed an algorithm that equalizes perceived contrast over the image, with some starting level of adapting luminance assumed. Contrast in the image is modified on the basis of the theoretical threshold-contrast curves of Heinemann (4). The workstation described herein extends the work of Liu and Nodine (8) to include adjustment of the input-output transfer characteristic for ambient illumination and for the gray-scale range of the particular mammogram being displayed. In this article, we describe the display station, development of the perceptually tempered display, and evaluation of the display station.
|
THE DISPLAY STATION
|
|---|
The display station shown in Figure 1 uses a computer (model GP6-266; Gateway 2000, Sioux City, Iowa) with a Pentium II processor (Intel, Santa Clara, Calif). The computer is interfaced to a gray-scale monitor (model DS5000L; Orwin Associates, Amityville, NY) by means of an interface board (model Md5/PCI-1; Dome Imaging Associates, Waltham, Mass). The computer software is written in IDL (Research Systems, Boulder, Colo), a high-level graphics language.
Before use of the display station, the video monitor was photometrically calibrated. A photometer (model J17; Tektronix, Beaverton, Ore) interfaced to the computer was used to measure the intensity of a 10 x 10-cm square of uniform luminance located in the center of the display surface. (The luminance of a display such as a CRT or a film illuminator is measured in foot-lamberts or candelas [cd] per square meter [1 foot-lambert = 3.4 cd/m2].) The intensity of the display surrounding the square was set at a luminance of 55 cd/m2, which was produced by a pixel driving intensity value of 128. The luminance was measured over 17 equally spaced pixel driving intensity values from 0 (black) to 255 (white); these pixel driving intensity values corresponded to a luminance of 1.7–343 cd/m2. Digitization and logarithmic transformation of the photometric data were performed; they were then displayed on the CRT along with an ideal perceptually linearized curve. The brightness and contrast controls were adjusted until the calibrated curve visually matched the ideal curve. Once the CRT is calibrated, it needs only occasional adjustment. The shape of the input-output transfer characteristic adjusted according to the perceptually linear display standard is shown in Figure 2 (top curve).
View larger version (29K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 2. Input-output transfer characteristic of the CRT (top curve) and final minimal detectable contrast (MDC) look-up table (bottom curve). The curves have a common pixel driving level axis. The nonlinearity of the MDC curve is exaggerated for purposes of illustration; the actual difference from the linear curve is usually more subtle. The effect of the MDC look-up table on the displayed image can be seen by following the dotted lines, which represent extrapolation from the image pixel value to the display luminance.
|
|
|
DEVELOPMENT OF THE PERCEPTUALLY TEMPERED DISPLAY
|
|---|
Estimation of the MDC
The MDC test pattern consists of nine horizontal bands of increasing intensity (Fig 3). Each band contains eight disks of decreasing contrast. This test pattern was displayed for each observer prior to a viewing session. The observer's task was to choose the "least visible" disk in each band. The observer's responses are affected by the display contrast and the ambient room lighting. A parabola was fitted to the contrast of each indicated disk and the intensity of the horizontal band; this parabola approximates the dependence of the observer's contrast sensitivity on display luminance at the level of ambient illumination (Fig 4).
View larger version (17K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 4. Approximation of the contrast sensitivity curve with a parabola. Heinemann (4) measured human contrast sensitivity at different levels of adapting luminance. Examples of this relationship at two adapting luminance levels are shown (solid lines). In reality, there is a whole family of curves of similar shape that have a minimum that shifts with the adapting luminance. Consider the lower curve, which corresponds to an adapting luminance of 100 cd/m2. The eye is maximally sensitive at a display luminance of 100 cd/m2, with an MDC of about 0.05. However, an object located in a dark part of the image at 10 cd/m2 would have to have a contrast of 0.1 to be seen. The practical solution in radiology is to use a spotlight to raise the luminance to 100 cd/m2 and improve the contrast sensitivity. As the adapting luminance decreases, the curves shift upward and maintain roughly the same shape. Attempts have been made to fit the curves from Heinemann's experimental data with simple equations (5). The algorithm of Liu and Nodine (8) required advanced information about adaptation level and was computationally intensive. We simplified that algorithm by assuming that a parabola (dashed lines) could be used to approximate contrast sensitivity at different levels of adapting luminance. The fit is reasonable at high adapting luminance (100 cd/m2), where radiologists prefer to operate. The fit for a lower adapting luminance (10 cd/m2 [upper curve]) is not very good. However, this luminance is well below a practical average viewing luminance.
|
|
Determination of the Range of Pixel Intensities of the Mammogram
As each case is displayed, the maximum and minimum pixel intensity in the breast parenchyma is determined by sampling over a region that includes breast tissue out to just beyond the skin line, thus excluding the extremes of pixel driving levels due to lead markers, labels, and cassette edge artifacts. Determination of the pixel intensity range is performed with a boundary detection procedure: After applying a median filter, an intensity threshold value 5% above the background (dark level) is selected. By means of this threshold, the breast image is transformed into a binary image and a contour is determined on the resultant image. Image intensities are then sampled on the original breast image along 30 equally spaced lines (Fig 5).
Production of the MDC Look-up Table
The best-fit parabola for MDC versus displayed luminance is integrated to produce an MDC-corrected look-up table. The maximum and minimum pixel driving levels determined from the mammogram are applied to the MDC-corrected look-up table so that the output intensity just matches the input intensity (Fig 2 [bottom half]). The MDC look-up table is designed to equalize the detectability of equal-contrast targets regardless of the regional mean pixel intensity surrounding the targets. The advantage of redistributing the contrast in this "tempered " fashion is to provide an initial view that allows visual access to the dark regions (fat, skin line) as well as the light regions (muscle, fibrous and ductal tissue). The viewers are still able to manipulate the gray scale of the image. All of the calculations and look-up table manipulations are done by using a 12-bit pixel intensity scale. This scale is transformed into an 8-bit scale for display.
Display of the Images
The CRT is photometrically calibrated as part of the regular quality assurance program. The MDC calibration is performed before each reading session with the ambient illumination set at 1.6 lux at the location of the observer's eyes. The calibration takes approximately 15–20 seconds to complete. The correction of each image is done off-line prior to the test. Observers are able to use a single slider to adjust the MDC look-up table. The slider can smoothly adjust the display from a look-up table, which produces the baseline perceptually linearized display standard, up to a maximum MDC setting. Figure 6a shows a breast image displayed with standard perceptually linearized display; Figure 6b shows the image displayed with MDC-tempered display, which allows visualization of the skin line.
View larger version (102K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 6a. Mammographic image displayed with standard perceptually linearized display (a) and MDC-tempered display (b). The skin line (arrow in b) is not visible in the standard perceptually linearized display (a).
|
|
View larger version (122K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 6b. Mammographic image displayed with standard perceptually linearized display (a) and MDC-tempered display (b). The skin line (arrow in b) is not visible in the standard perceptually linearized display (a).
|
|
|
EVALUATION OF THE DISPLAY STATION
|
|---|
Our development cycle includes periodic benchmark testing by using a sample of cases from a digital database of normal and abnormal mammograms, in which all of the malignancies and many of the benign lesions are histologically proved. The mammograms were originally obtained on film and were digitized to a pixel size of 100 mm with a digitizer (Lumiscan 100; Lumisys, Sunnyvale, Calif). Readers are shown a craniocaudal view and a mediolateral oblique view and are asked to move a pointer on the display to any potential malignant lesion and click the mouse. The response time from the start of viewing each case and the location of the pointer are recorded by the software. After the click, a pull-down menu appears; the reader must select one or more diagnoses (ie, mass, calcification, or architectural distortion) and indicate a confidence level for malignancy. These data are used to compute a receiver operating characteristic (ROC) curve and determine the area under the curve.
Two mammographers (S.P.W., E.F.C.) and a general radiologist (H.L.K.) were tested on 75 mammograms: 25 with malignancies, 25 with benign lesions, and 25 that were normal. Table 1 is a comparison of the areas under the ROC curve. Although each reader did better with the MDC-tempered display, the difference was not significant when tested with a paired t test. The time to the first pointing out of a lesion was very variable but on average was not different for the two display modes (Table 2).
|
CONCLUSIONS
|
|---|
The accuracy and speed of the tempered display function are equal to those of the standard perceptually linearized display function when used on a moderately bright monitor (300 cd/m2). With the tempered display function, the initial view of the image provides visual access to lighter and darker regions of display with some sacrifice of visual access to medium-intensity regions. The display can be adjusted by moving a single slider, which is an attempt to simplify the user interface. Development of the display station is continuing with the addition of the use of verbal commands to modify display parameters and an eye position–contingent roving window.
|
Footnotes
|
|---|
Abbreviations: CRT = cathode ray tube
MDC = minimal detectable contrast
ROC = receiver operating characteristic
|
References
|
|---|
-
Blume H, Roehrig H, Browne M, Ji TL. Comparison of the physical performance of high resolution CRT displays and films recorded by laser image printers and displayed on light-boxes and the need for a display standard. Proc SPIE 1990; 1232:97-114.
-
Johnston RE, Zimmerman JB, Rodgers DC, Pizer SM. Perceptual standardization. Proc SPIE 1985; 536:44-49.
-
Blume H, Hemminger BM. Image presentation in digital radiology: perspectives on the emerging DICOM display function standard and its application. RadioGraphics 1997; 17:769-777.[Abstract]
-
Heinemann E. The relation of apparent brightness to the threshold for differences in luminance. J Exp Psychol 1961; 61:389-399.[Medline]
-
Cobra D. Image histogram modification based on a new model of visual system nonlinearity. J Electron Imaging 1998; 7:807-815.
-
Pizer SM, Amburn EP, Austin JD, et al. Adaptive histogram equalization and its variations. Comput Vision Graph Image Process 1987; 39:355-368.
-
Mokrane A. A new image contrast enhancement technique based on a contrast discrimination model. Comput Vision Graph Image Process 1992; 54:171-180.
-
Liu H, Nodine CF. A generalized image contrast enhancement technique based on the Heinemann contrast discrimination model. J Electron Imaging 1996; 5:388-395.
This article has been cited by other articles:
|
|
|
|
 
H. L. Kundel, C. F. Nodine, E. F. Conant, and S. P. Weinstein
Holistic Component of Image Perception in Mammogram Interpretation: Gaze-tracking Study
Radiology,
February 1, 2007;
242(2):
396 - 402.
[Abstract]
[Full Text]
[PDF]
|
|
|
|
|
|
|
C. F. Nodine, C. Mello-Thoms, S. P. Weinstein, H. L. Kundel, E. F. Conant, R. E. Heller-Savoy, S. E. Rowlings, and J. A. Birnbaum
Blinded Review of Retrospectively Visible Unreported Breast Cancers: An Eye-Position Analysis
Radiology,
October 1, 2001;
221(1):
122 - 129.
[Abstract]
[Full Text]
[PDF]
|
|
|
Top
Abstract
INTRODUCTION
