Full-Field Digital Mammography on LCD Versus CRT Monitors

LCD Versus CRT Monitors for Mammography Women s Imaging Original Research WOMEN S IMAGING Margarita L. Zuley 1 Kathleen M. Willison 1 Ermelinda Bonaccio 2 David P. Miller 3 David L. Leong 4,5,6 Posy J. Seifert 1 Patricia Somerville 1 Stamatia Destounis 1 Zuley ML, Willison KM, Bonaccio E, et al. Keywords: breast, breast cancer, mammography, PACS DOI:10.2214/AJR.04.1713 Received November 5, 2004; accepted after revision September 18, 2005. 1 Department Radiology, The Elizabeth Wende Breast Clinic, 170 Sawgrass Dr., Rochester, NY 14620. Address correspondence to M. L. Zuley. 2 Department Radiology, Roswell Park Cancer Institute, Buffalo, NY. 3 Ovation Research Group, Highland Park, IL. 4 Digital Radiography Systems Division, Analogic Corporation, Peabody, MA. 5 UCD School Medicine and Medical Sciences, Dublin, Ireland. 6 Department Mathematics and Statistics, University New Hampshire, Durham, NH. AJR 2006; 187:1492 1498 0361 803X/06/1876 1492 American Roentgen Ray Society Full-Field Digital Mammography on LCD Versus CRT Monitors OBJECTIVE. Our purpose was to determine if the display full-field digital mammograms on a 5-megapixel liquid crystal display (LCD) monitor is at least equivalent to the display the same on a 5-megapixel cathode ray tube (CRT) monitor. MATERIALS AND METHODS. Five radiologists evaluated normal anatomy and features 61 abnormalities in 48 full-field digital mammograms. A 9-point Likert scale was used to compare images on two identical st-copy review workstations, one equipped with two 5- megapixel CRTs and the other with two 5-megapixel LCDs. Outcomes were evaluated using a random-effects analysis variance model. Means and SEs were reported. Ninety-five percent confidence intervals and p values were calculated. RESULTS. The two systems were equivalent for most features. The LCDs were rated better for the sharpness mass margins (p = 0.011) and mass conspicuity (p = 0.050). For calcium features, the LCDs were rated better than the CRTs for lesion conspicuity (p = 0.010) and number calcifications (p = 0.043). For architectural distortions, there was no statistically significant difference between the monitors in any the features evaluated. For display characteristics, the LCDs were better for luminance (p = 0.021). The CRTs were significantly better for image noise (p = 0.001). In the overall ratings, there was no statistically significant difference between the two displays. CONCLUSION. The 5-megapixel monochrome active-matrix LCD is equivalent to and in some respects better than the 5-megapixel CRT display for full-field digital mammograms over a range normal and abnormal findings. he widespread implementation T full-field digital mammography has naturally raised questions about the optimal display, storage, and retrieval images. Today, primary interpretation digital mammography is typically performed on the st-copy review workstation, which is purchased with the acquisition unit. Integral in the approved st-copy review workstation are two 5-megapixel (MP) (2,048 2,560 pixels) cathode ray tube (CRT) monitors. Simultaneously, the rest radiology is moving away from CRT monitors for several reasons. CRTs typically have a low luminance (300 cd/m 2 ), requiring ambient light levels to be so low that the inevitable comparison with prior studies on an alternator or viewbox is difficult because the average luminance from alternators or viewboxes is approximately 2,000 cd/m 2. CRTs also have a short life expectancy (with slow degradation display quality over time) approximately 36 months (about 30,000 hours), requiring replacement rather than repair. For optimal function in a standard reviewing room, CRTs are typically set at their highest luminance levels, which further decreases the life the monitor. Eye fatigue is also a problem because CRTs require the screen to be constantly refreshed, or repainted, because the image on the screen relies on light emitted from phosphors that fade quickly. Despite improvements in refresh-rate standards from 60 to 75 MHz by Video Electronics Standards Association (VESA) standards, eye fatigue is still an issue for the radiologist. In addition, the emissive nature the CRT results in blooming the focused beam at the periphery the monitor, which degrades resolution in these areas. This emissive nature also creates veiling glare (unwanted scattered light), which results in degradation the overall contrast resolution. CRTs are heavy ( 40 lb [18 kg] each); they have a large, cumbersome footprint; and they have a high heat output, which requires additional air conditioning. For these and other reasons, liquid crystal display (LCD) monitors are coming into favor. 1492 AJR:187, December 2006

LCD Versus CRT Monitors for Mammography TABLE 1: Case Distribution s Mass Cases Calcium Cases Architectural Distortion Cases Margin Type Type Well circumscribed 6 Skin 1 Lucent center 5 Microlobulated 4 Vascular 1 Dense center 5 Indistinct 8 Coarse 1 10 Obscured 4 Round 0 Spiculated 8 Lucent-centered 0 30 Rim 3 Milk calcium 2 Density Dystrophic 0 High density 6 Punctate 3 Isodense 17 Amorphous 5 Low density 6 Pleomorphic 4 Containing fat 1 Fine linear branching 1 30 21 Distribution Grouped (clustered) 11 Linear 1 Segmental 4 Regional 3 Diffuse 2 21 LCD monitors are lightweight (< 15 lb [7 kg]), have a small footprint, and are becoming less cost-prohibitive. They have a long life expectancy, with minimal cost impact replacing the fluorescent lamp backlight. Refresh rates are not a concern with LCD monitors because they virtually hold a charge until updated. The design the LCD does not require a focused beam to produce an image. Instead, an electronic current is applied to a thin-film transistor (TFT), rendering the entire surface the LCD a uniform resolution. Moreover, the flat-panel design provides a better overall resolution for a TABLE 2: Lesion Size and Distribution Breast Densities given display matrix as illustrated in a study comparing the clinical impact 3-MP LCD with 5-MP CRT for lung nodules (Siegel E et al., presented at the 2002 annual meeting the International Society for Optical Engineering). Monitor luminance has been shown to be at least as important as monitor resolution [1 3], and with superior luminance approximately 700 cd/m 2 nearly double the luminance the CRT LCDs are an attractive alternative to CRTs. To date, some studies have shown equivalence between LCDs and CRTs in the display radiographic abnormalities [4]. Our study was designed to determine if the display full-field digital mammograms on a 5-MP LCD monitor was at least equivalent to the display the same on a 5-MP CRT monitor. This study was performed in the context trying to determine the optimal display fullfield digital mammography images and represents just one many small steps necessary to be able to integrate the display, storage, and retrieval digital mammography into PACS systems, which so many us already use. Materials and Methods Study Design The study involved a comparison mammographic and display features viewed on two identical st-copy review workstations using a 9-point Likert scale. On one these workstations, reviewers used two 5-MP CRTs; on the other workstation, reviewers used two 5-MP active matrix LCDs. Case Selection The reports all screening mammograms performed on one full-field digital mammography unit (SenoScan, Fischer Imaging) from March 2003 to November 2003 were reviewed (n = 2,500). The hardcopy images all cases that were given a BI-RADS category 2 or 0 were evaluated (n =331) by the study coordinator in medical record number order to obtain as random a sample as possible. We then enriched the data set with missing lesion types and breast densities. This was done with the intention representing the spectrum and frequency mammographic abnormalities to include mass; calcium; mass with calcium; and architectural distortion in dense, heterogeneously dense, scattered, and adipose-replaced tissue types. The resultant data set consisted 48 cases containing 61 abnormalities including 30 mass lesions, 21 calcium lesions, and 10 architectural distortions. Asymmetric densities were grouped in the mass category. Two the 48 cases had a calcified mass. The calcium and mass features for these two cases were evaluated separately by the reviewers and the results included in the mass and calcium numbers. Tables 1 and 2 show the distribution Size (cm) Breast Density Heterogeneously Lesion Type Lesions Mean SD Min Max Fatty Fibratty Dense Dense Studies Mass 30 1.4 0.15 0.42 4.1 5 9 7 2 23 Calcifications a 21 1.4 1.4 0.3 5.6 1 8 5 2 16 Architectural distortion 10 1.3 0.4 0.6 2.3 1 4 4 0 9 Average 1.4 0.7 0.4 4.0 61 7 21 16 4 48 Note Breast density given per case. Min = minimum, Max = maximum. a Size range individual calcifications, < 0.01 9 mm. AJR:187, December 2006 1493

Fig. 1 Images were viewed on paired monitors, with 5-megapixel (MP) liquid crystal display (LCD) monitors in center immediately adjacent to each other and 5-MP cathode ray tube (CRT) monitors flanking LCDs. Note that image difference is photographic effect, not an indication monitor capacity or limitations. lesion features, breast density, and size. In addition, five normal cases were chosen across the four tissue types. Scale Development A 9-point, non forced-choice Likert scale was designed to evaluate the anatomic and pathologic mammographic features listed in Appendix 1 and the thickness and conspicuity the skin line and subcutaneous tissue. St-copy display characteristics, including luminance, dynamic range, sharpness, background homogeneity, image distortion, display noise, image size, and image noise, were also included in the Likert scale. For computing a mean for each characteristic, the 9-point Likert scale was constructed with text anchors indicating that each end the scale corresponded to one monitor system being 100% better (dramatically better) than the other and the middle value corresponded to equality. The other intermediary values were 75% better (significantly better), 50% better (moderately better), and 25% better (slightly better) in one direction or the other. Thus, the completed 1 9 Likert scale was easily transformed to a comparative performance scale negative 100% to positive 100%, in which positive values indicate a preference for the LCD and negative values indicate a preference for the CRT. From this, the mean preference the reviewers for each characteristic was computed. Positive or negative values were not assigned to either system during reviewer sessions. The Likert scale design was based on the mammographic features detailed and defined by the American College Radiology BI-RADS [5], which is familiar to all radiologists and quite complete in detailing mammographically found features. Five test cases were then reviewed by two board-certified radiologists who are fellowship trained in mammography. Based on this test cohort, the scale was adjusted for the full study group analysis. Reviewer Training Before starting the evaluations, the three radiologists who were not involved in the Likert scale development participated in a training session to become familiar with the scale, the definitions, and the image display protocols. The test set five mammograms that were initially used to refine the Likert scale was also used for this reviewer training session. Case Analysis The study group was evaluated independently by the five radiologists. The reviewers had an average 8 years (range, 5 12 years) experience in screen-film mammography and 1 year experience in st-copy reviewing digital mammography. Each radiologist interprets approximately 30,000 mammograms per year in our practice. The evaluations were done with the cases displayed in random order as to tissue type and abnormality. Because the study was not done to evaluate each radiologist s interpretation skills, and to avoid the possibility inadvertent evaluation the wrong lesion, the radiologist was directed to the lesion interest with a lesion-specific data form that indicated the type and geographic location the lesion interest. Four mass features were assessed: shape, margin sharpness, density, and conspicuity. The calcium features analyzed were number, shape, and sharpness edges; distribution; and conspicuity. Four architectural distortion features were evaluated: spiculation, density, parenchymal edge distortion, and conspicuity. For all cases, including the five normal cases, the skin was evaluated for thickness, subcutaneous tissue visibility, and overall conspicuity. In addition, display characteristics, including luminance, dynamic range, image sharpness, background homogeneity, image distortion, display noise, and image size, were compared (see Appendix 2 for definitions). Last, for each mammogram, the reviewers were asked to provide an overall assessment the CRTs and LCDs from zero to 100% better for either monitor, with zero being equivalent. This overall assessment was not tied to any word anchors, allowing the reviewer to choose from the full range the scale. The Monitors Two identical st-copy review workstations were used to conduct the study. One was equipped with and optimized for two 5-MP CRTs (MGD521M, Barco) with Dome R5 (5-MP) display controllers (Planar Systems), and the other was equipped with and optimized for two 5-MP monochrome active-matrix LCDs (Dome C5i with Dome DX [5-MP] display controllers, Planar Systems). Images were sent simultaneously to both workstations for display. Both monitor systems were set to accept 8-bit images and display at 8 bits. Target monitor luminance was chosen by the vendor and calibrated as would be typical in the clinical setting, with the CRT pair set at 300 cd/m 2, and the LCD pair set at 550 cd/m 2. Before each session, 1494 AJR:187, December 2006

LCD Versus CRT Monitors for Mammography quality assurance testing was performed as specified by the vendor. Full-Field Digital Mammograms The full-field digital mammography images were acquired using the SenoScan. This system acquires images at 12 bits and then performs a 12- to 8-bit transformation on the data before sending the images to the st-copy review workstation for display. This process is standard for this unit. The acquisition and display bit depth were not altered for this study. Data Collection Two 1-hour reviewing sessions were arranged for each radiologist to evaluate the 53 cases (five normal and 48 with abnormalities). The reviewers were blinded to the other reviewers results. During each session, the research assistant filled in all the data on the Likert scales. The cases were reviewed by acquisition-date order so that they were random with respect to tissue and lesion type. The images were viewed on paired monitors, with the 5-MP LCDs in the center immediately adjacent to each other and the 5-MP CRTs flanking the LCDs (Fig. 1). The radiologists had the ability to put any pair monitors in black-screen mode if the light was limiting evaluation the other pair. The images were always first displayed with identical window and level parameters on both sets monitors; however, the radiologist could change the window and level any the images during the study. The cases were interpreted in a standard reviewing room without ambient light. A definition and guideline sheet each item to be assessed was prepared and provided to the reviewers (Appendix 2). Once the session was over, the radiologist could not change any responses. Hanging Protocol The hanging protocols between comparison components were matched. The cases were displayed on each set monitors with the following hanging protocol: four-view mammogram, four views were hung on one monitor; bilateral craniocaudal (CC) views were hung simultaneously, one each on the right and left monitor each set monitors followed by bilateral mediolateral oblique (MLO) images displayed in the same way. Finally, the two views each breast (CC and MLO views) were displayed again at full resolution, one on each monitor. The monitors were all turned slightly toward the midline so that the reviewer had as close to a perpendicular viewing angle to the face each monitor as possible. Statistical Analysis All outcomes were evaluated using a random effects analysis variance model in which both reviewer and case are treated as random effects. The advantage this model is that the p values and confidence intervals are calculated in a way that takes into account the likely clustering ratings for the same case reviewed by multiple reviewers and the likely clustering ratings for the same reviewers in their reviews multiple cases. In this way, the results may be projected to a new reviewer interpreting a new case. For each outcome, the mean and model-based SEs were reported. A 95% confidence interval and the p value were calculated for each characteristic for which all the reviewers did not rate the monitors as equivalent. Interobserver variability for the five reviewers was reported when the difference between systems was statistically significant. In these instances, the range individual reviewer averages was reported. Intraobserver variability could not be truly evaluated because each reviewer interpreted each case only once. However, intraobserver variability was addressed by presenting ranges individual case averages across reviewers. This is the case variability, which is defined as the average preference that the reviewers had for each case given any particular feature. This is important to illustrate that the results were not skewed by any one case but also shows the extent to which the level preference did differ over cases. Results Mass s The LCD and CRT displays were both statistically and clinically equivalent with respect to the ratings most mass features (Table 3). The largest difference in display was present for margin sharpness, which was rated on average 3.6% better on the LCD monitors (p = 0.011), with individual reviewer averages for the five reviewers ranging from 0.8% better to 8.6% better. The reviewers preferred the LCDs for mass conspicuity (1.3% on average; variation in reviewer averages, 0.8 1.7%; variation in case averages, 5% to 25%), but the preference was on the border being statistically significant (p = 0.05) because the upper and lower bounds the 95% confidence limits suggest that the difference is unlikely to be more than 5% favoring one monitor or the other. Calcium s The LCD monitors were either equivalent or better than the CRTs with respect to calcium features (Table 4). In particular, reviewers favored the LCD for conspicuity (6.2% better, p = 0.010; variation in reviewer averages, 0 10.7% better; variation in case averages, 5.0% worse to 25.0% better). Reviewers also favored the LCDs for number calcifications (2.4%, p = 0.043; variation in reviewer averages, 0 6.0% better; variation in case averages, 0 15.0% better). The LCD and CRT displays did not differ with respect to shape, sharpness edges, or distribution calcifications. Architectural Distortion s The two displays did not differ significantly for architectural distortion features (Table 5). The observed differences favoring the LCD system fell well short statistical significance (p = 0.588 for spiculation, p = 0.323 for parenchymal edge distortion, and p = 0.802 for conspicuity). A rating no difference was given by every reviewer for every case for the density feature architectural distortion, so no SE or inferential statistics could be computed; however, the consistent rating zero is clearly strong evidence in favor equivalence. Evaluation Display s for All Cases, Including the Normal Set In the evaluation display features, which included both the normal cases and the cases containing abnormalities (Table 6), the LCDs were significantly better with respect to luminance (14.3%, p = 0.021). Individual reviewer averages ranged from 2.2% to 26.5% better for luminance. The CRTs had a significant advantage for image noise (2.8%, p < 0.001; variation in reviewer averages, 0.5 4.8% better; variation in case averages, 0 10% better). The two displays did not significantly differ for dynamic range, skin thickness, subcutaneous tissue, conspicuity the skin and subcutaneous tissues, image distortion, display noise, or image size. Overall In addition to the feature-by-feature rating, an overall rating was provided on a continuous 100% to 100% scale (Table 7). The LCD and CRT displays were both statistically and clinically equivalent for evaluation all features included. Discussion The demanding nature mammographic imaging has delayed its entry into the digital age. Even with the recent regulatory approval full-field digital mammography units, the road to digital mammography has been fraught with transitional issues, not the least which is the cost implementing a digital mammography program. However, the initial cost the full-field digital mammography acquisition AJR:187, December 2006 1495

TABLE 3: Results Comparison for Mass p Shape 149 0.50 0.374 0.24 1.24 0.181 Margin sharpness 149 3.55 1.374 0.83 6.26 0.011 Density 149 0.67 0.465 0.25 1.59 0.152 Conspicuity 149 1.34 0.678 0.00 2.68 0.050 Note Ratings = number lesions number reviewer interpretations recorded. Mean is average preference all reviewers by percentage. Positive numbers indicate preference for the liquid crystal display and negative numbers indicate a preference for the cathode ray tube. CL = confidence limits. TABLE 4: Results Comparison for Calcium p Number 105 2.38 1.163 0.07 4.69 0.043 Shape 105 1.43 1.320 1.19 4.05 0.282 Sharpness edges 105 7.14 4.206 1.20 15.48 0.092 Distribution 105 0.71 0.476 0.23 1.66 0.137 Conspicuity 105 6.19 2.363 1.50 10.88 0.010 Note Ratings = number lesions number reviewer interpretations recorded. Mean is average preference all reviewers by percentage. Positive numbers indicate preference for the liquid crystal display and negative numbers indicate a preference for the cathode ray tube. CL = confidence limits. TABLE 5: Results Comparison for Architectural Distortion p Spiculation 47 1.02 1.874 2.75 4.79 0.588 Density 47 0.00 0.000 Parenchymal edge distortion 47 2.21 2.212 6.66 2.24 0.323 Conspicuity 47 0.47 1.856 3.27 4.20 0.802 Note Ratings = number lesions number reviewer interpretations recorded. Mean is average preference all reviewers by percentage. Positive numbers indicate preference for the liquid crystal display and negative numbers indicate a preference for the cathode ray tube. Dash ( ) indicates not applicable. CL = confidence limits. unit is just one aspect. In order for digital mammography to be included in whole-department PACS systems, issues such as archiving, universal st-copy review workstations, and monitor choices are all critical. As directed by the United States Food and Drug Administration, each vendor full-field digital mammography units has provided a package for digital mammography from acquisition to display to storage. As we move out the research arena with digital mammography and into the high-volume use this technology, it is imperative that we work to make this technique fit into our preexisting PACS systems. In doing so, we will help control the overall costs the systems both in dollars spent and in interpretation time. Because LCD monitors are now widely used for the primary interpretation other techniques, it seemed logical to evaluate their use for mammography. This study showed that the 5-MP LCD display is equivalent to and in some respects better than the 5-MP CRT display for full-field digital mammograms over a range mammography cases. We found that the LCD monitors showed improved calcium conspicuity. Likely, these results are at least in part due to the fixed matrix the LCDs. This fixed matrix creates a crisper image compared with a CRT because each pixel in the LCD matrix remains constant over time in both location and size. In contrast, each pixel in the matrix the CRT varies over time in both location and size, producing an inherent slight blurring pixel edges. The light source the CRT is fired thousands times per second to create the resultant image. This constant refreshing each pixel is minutely variable in location rather than fixed. Further, because the technology is based on a fired light source, there is slight di- vergence the light beam as it travels through space, producing a slight blur. These two aspects the CRT result in a slightly smoother display than the LCD, but one that is less sharp. The fixed matrix the LCD produces a more grainy display. Image noise was the only characteristic that was found to be statistically superior on the CRT. However, in our study this image noise, or graininess, did not affect the radiologists ability to evaluate lesion features. The structured noise inherent in the design the TFTs likely accounts for the noise seen on the LCDs in this study. Newer TFT designs are eliminating this problem. Background homogeneity was also rated superior on the CRTs but fell short statistical significance. The reduction background homogeneity on the LCDs is due to the limitations the viewing angle and luminance fallf from f-axis viewing with the LCDs. One way to limit this problem is to carefully maintain a viewing angle as close to 0 as possible (straight-on viewing or perpendicular to the screen). Further, CRTs typically have a blacker background than LCDs. It is our opinion that the increased luminance and wider dynamic range the LCDs more than compensate for this difference and that this difference did not affect our ability to assess lesion features. Among the lesion features for which differences were identified between systems, there was limited interreviewer variability. In these instances, all five reviewer averages showed either no preference or a preference for the same system. Case variability was somewhat greater. However, there was no outlier case that skewed the data in favor one system or another for any the features. For example, the CRT was not preferred for any single case or by any reviewer for number calcifications, and the LCD was not preferred for any single case or by any reviewer for image noise. Finally, in the overall assessment the two monitors, no statistically significant difference was noted. There are several limitations this study. The comparison was done with an 8-bit display system using images that are intended to be displayed at 8 bits. Some full-field digital mammography manufacturers display at 10 bits. The industry standard for display digital radiography has historically been set at 8 bits based on research showing that the human eye can realistically see approximately 256 shades gray (or 8 bits). The monitors evaluated in this study have an 8-bit control card and 8-bit display. It has not yet been 1496 AJR:187, December 2006

LCD Versus CRT Monitors for Mammography TABLE 6: Results Display Comparison for All Cases, Including Normal Cases p Thickness 263 0.10 0.095 0.09 0.28 0.318 Subcutaneous tissue 263 0.10 0.095 0.09 0.28 0.318 Conspicuity 263 0.10 0.095 0.09 0.28 0.318 Luminance 264 14.33 6.165 2.19 26.47 0.021 Dynamic range 264 7.61 5.095 2.42 17.64 0.137 Sharpness 264 4.13 4.138 4.02 12.28 0.319 Background homogeneity 264 13.48 14.335 41.71 14.74 0.348 Image distortion 264 0.09 0.095 0.09 0.28 0.320 Display noise 264 0.47 0.599 0.71 1.65 0.432 Image size 264 0.00 0.000 Image noise 255 2.77 0.820 4.38 0.15 0.001 Note Ratings = number lesions number reviewer interpretations recorded. Mean is average preference all reviewers by percentage. Positive numbers indicate preference for the liquid crystal display and negative numbers indicate a preference for the cathode ray tube. Dash ( ) indicates not applicable. CL = confidence limits. TABLE 7: Results Overall Rating Stratified by Primary Lesion TABLE 1: Results Type Overall Rating Stratified by Primary Lesion Type p Mass 125 1.07 3.505 8.01 5.87 0.761 Calcium 85 2.82 2.558 2.27 7.91 0.273 Architectural distortion 39 1.66 3.184 8.11 4.78 0.605 Note Ratings = number lesions number reviewer interpretations recorded. Mean is average preference all reviewers by percentage. Positive numbers indicate preference for the liquid crystal display and negative numbers indicate a preference for the cathode ray tube. CL = confidence limits. shown that 8-bit display is the most optimal bit-depth display for digital mammography. In fact, there is some controversy among digital mammography manufacturers regarding the optimal bit depth for acquisition or display digital mammograms. Full-field digital mammography images are typically acquired at a predetermined bit depth (12 14 bits) and then undergo processing to transform the images to 8 12 bits for transfer. Finally, the images that arrive at the st-copy review workstation may require that the display control card perform another transformation for final display at 8 10 bits. Users should understand the acquisition bit depth, the hardware that performs this processing, and the display bit depth the system they are using. The work to determine the optimal bit depth for display full-field digital mammography still needs to be done. The physical setup placing the LCDs in the center flanked by the CRTs may also have presented a potential bias. This setup was chosen to eliminate problems with f-angle viewing on the LCDs. We tested the arrangement the LCDs flanking the CRTs before starting the study. The luminance fallf and rainbow effect resulting from this arrangement caused a significant loss in perceptible image quality the LCDs, so this arrangement was not used in the trial. Conversely, because all the reviewers had significant previous experience interpreting full-field digital mammograms on CRTs, a potential bias existed toward that with which the radiologists were already comfortable, namely, the CRTs. Unfortunately, in this study design, it is impossible to eliminate all bias because it is readily apparent which monitor is which. We considered these potential biases unavoidable for a direct side-by-side comparison the images. An alternative design would have been to evaluate the monitors in different reviewing sessions, but because we were using the CRTs as the gold standard and the details mammographic lesions are so subtle, we thought the most accurate comparison would be side by side so that these subtle differences could be detected and evaluated. In summary, we found that 5-MP flat-panel monitors are at least equivalent to and in some aspects superior to 5-MP CRTs in the display full-field digital mammographic images. References 1. Herron JM, Bender TM, Campbell WL, Sumkin JH, Rockette HE, Gur D. Effects luminance and resolution on observer performance with chest radiographs. Radiology 2000; 215:169 174 2. Ikeda M, Ishigaki T, Shimamoto K, et al. Influence monitor luminance change on observer performance for detection abnormalities depicted on chest radiographs. Invest Radiol 2003; 38:57 63 3. Kimme-Smith C, Haus AG, DeBruhl N, Basset LW. Effects ambient light and view box luminance on the detection calcifications in mammography. AJR 1997; 168:775 778 4. Krupinski EA, Johnson J, Roehrig H, Nafziger J, Fan J, Lubin J. Use a human visual system model to predict observer performance with CRT vs LCD display images. J Digit Imaging 2004; 17:258 263 5. American College Radiology. Breast imaging reporting and data system atlas (BI-RADS mammography), 4th ed. Reston, VA: American College Radiology, 2003 Appendixes appear on next page AJR:187, December 2006 1497

APPENDIX 1: Mammographic s Evaluated on the Likert Scale Abnormality Mass Calcium Architectural distortion Mass with calcifications Shape Margin sharpness Density Conspicuity Number Shape Sharpness edges Distribution Conspicuity Spiculation Density Parenchymal edge distortion Overall conspicuity s both mass and calcium APPENDIX 2: Definitions and Guidelines Used in the Likert Scale Evaluation Display Characteristic Luminance Dynamic range Sharpness Background homogeneity Image distortion Display noise Image size Image noise Guideline Maximum brightness the display. Does the image appear brighter on one monitor? amount gray-scale information that is visible on the display. Do you see more shades gray on one monitor versus the other? Picking a very dark, very bright, and midgray location, look at the same location an image on both monitors. Is the difference in minimum, maximum, and midgray noticeable? Remember that the very bright area is the same as luminance. Edge detail or crispness the image on the display. Are the edges objects easier to see? Are calcifications better defined on one monitor? Look at the image in the center and at the corners the image. Defined as the perception color change (rainbow-like ripple) evident in the background the monitor. Geometric nonuniformity in the display. Does the image appear to be distorted? Do straight lines appear curved? Look at the image in the center and at the corners the image. Noise that is seen in the displayed image that will vary with time. Does the image appear noisier in one monitor versus the other? Look at the image in the center and at the corners the image. Display noise changes with time; X-ray quantum noise will not change. Wavy lines or moving streaks are considered noise. The physical size the displayed image. Does the display size on one monitor versus the other affect your ability to detect and classify masses, calcifications, and architectural distortions? An overall grainy appearance throughout the image on the liquid crystal display (LCD) monitors. Does the structured noise in the LCD display affect your ability to assess the image and lesion features? 1498 AJR:187, December 2006