Color CRT displays need work for primary interpretation

The transition to filmless radiology and PACS has revolutionized the way in which radiographic images are viewed and interpreted. Viewboxes are becoming obsolete, and monitors are becoming the predominant viewing medium. Given the variety of monitors available, the question becomes, What displays are best suited for use in radiology?

Our research laboratory approaches this question from an image perception perspective. One issue that affects the choice of a monitor is the suitability of off-the-shelf color monitors for primary interpretation. Radiologic images are intrinsically rendered in shades of gray, so monochrome monitors would seem to be the appropriate choice for display, even though evidence suggests that humans can discriminate more levels of color than of gray.(1) High-resolution monochrome cathode ray tube (CRT) displays with the appropriate video boards are rather expensive, however, since the market for them is limited. Color displays are much more affordable, but are they appropriate for use in radiology?

In some circumstances, color displays are certainly desirable and appropriate. Doppler ultrasound relies on color rendering in the display to highlight changes in blood flow and other dynamic information. The same is often true in SPECT and PET imaging. Color displays are acceptable for these applications, in part because the resolution requirements are not that demanding. A typical color CRT is 1600 x 1200 pixels, which is quite suitable for these modalities, in which high resolution is not generally an issue. Color additions are also useful for applications involving image registration, such as PET with MRI data, and volume rendering. But suitability is more ambiguous with modalities that do require higher resolution images, such as computed radiography and mammography.

LUMINANCE AND SPATIAL RESOLUTION

Compared with monochrome displays, color displays have a lower luminance (30 to 40 foot lamberts compared with greater than 200 foot lamberts), which can negatively affect contrast perception, and reduced spatial resolution, which affects the perception of fine details and textural differences.(2) The question we sought to answer was whether these differences would translate into an effect on detection performance with radiographic images.

We conducted a study(3) that showed a series of chest images with subtle solitary pulmonary nodules to six radiologists. The images were displayed once on a monochrome CRT monitor (Image Systems, Minnetonka, MN) and once on a color CRT monitor (Iiyama, Nagano, Japan). Both monitors were in the midresolution range (1600 x 1200 pixels) since color CRTs are not readily available in higher resolutions off the shelf. The radiologists were instructed to search the displayed images for nodules and report their decision and confidence. Their eye position was recorded(4) as they searched the images using an eye and head tracker (model 4000SU, Applied Science Laboratories, Bedford, MA).

Diagnostic accuracy was measured using multireader multicase receiver operating characteristic (MRMC ROC) analysis for determining area under the curve (Az) values. All six readers had higher performance with the monochrome than with the color CRT display (Figure 1). The average Az was 0.9235 for the monochrome monitor and 0.7769 for the color monitor (p

The eye-position recording also revealed significant differences between the monochrome and color monitors in terms of specific search parameters. Figure 2 shows a typical pattern of one of the radiologists searching for pulmonary nodules. Overall search time was significantly longer with the color monitor: 42.61 seconds on average versus 33.96 seconds with the monochrome monitor. Ten seconds may not seem that long, but compounded by the total number of images a radiologist is likely to view in a given day, the total extra viewing time can be significant.

The second search parameter that was affected by the type of monitor used to display the image was time to first fixate or have the axis of gaze fall on the nodule during the search. With the monochrome monitor, the average time to first fixate nodules when they were reported (true positives) was 415.17 msec compared with 526.24 msec with the color monitor. For lesions that were fixated but not reported (false negatives), time to first fixate was 446.82 msec with the monochrome and 602.04 msec with the color monitor. For both decisions, the time to first fixate the nodules was significantly longer with the color monitor.

Finally, we analyzed the dwell times associated with each type of decision: true and false, positive or negative (Figure 3). For true positive (TP), true negative (TN), and false positive (FP) decisions, the median dwell times were all longer for color than for monochrome viewing, especially for the two types of positive decisions. The radiologists took significantly longer to render a positive decision (right or wrong) with the color monitor, suggesting that more visual processing was required to extract the relevant information from the images and reach a decision with the color display. Only the false-negative (FN) decisions were longer with the monochrome monitor. This suggests that the radiologists may have been able to detect more nodule features, thus attracting more visual scrutiny, with the monochrome, even though they were still not able to recognize those features as belonging to a true nodule.

COLOR UNACCEPTABLE

The results of our study indicate that viewing radiographic images on a color CRT is not acceptable, at least for the task of searching for subtle pulmonary nodules in chest images. Detection performance and visual search efficiency were both significantly lower with the color display. It is likely that the results apply as well to other types of high-resolution images, such as bone and mammographic images, that require fine detail perception.

We did not conclude that color displays will never be appropriate. Advances in liquid crystal display (LCD) are occurring daily. LCDs have certain advantages over the traditional CRT monitors that may make color feasible, and they can achieve higher luminance levels, reducing the impact on contrast perception. The manner in which color is rendered with LCD differs considerably from that used with CRT displays, making the effect on spatial resolution significantly less important. Further testing, of course, needs to be done as these new display technologies are introduced into radiology.
DR. KRUPINSKI is a research associate professor of radiology at the University of Arizona in Tucson.

References
1. Levkowitz H, Herman GT. Color scales for image data. IEEE Comp Graphics and Applic 1992;12:72-80.
2. Behlen FM, Hemminger BM, Horii SC. Displays. In: Kim Y, Horii SC, eds. Handbook of Medical Imaging Volume 3 Display and PACS. Bellingham, WA: SPIE Press, 2000:403-440.
3. Krupinski EA, Roehrig H. Pulmonary nodule detection and visual search: P45 and P104 monochrome versus color monitor displays. Acad Radiol 2002;9:638-645.
4. Nodine CF, Kundel HL, Toto LC, Krupinski EA. Recording and analyzing eye-position data using a microcomputer workstation. Behav Res Meth , Instrum, & Comp 1992;24:475-485.