| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Thoracic Imaging |
1 From the Department of Radiology and Ludwig Boltzmann-Institute for Clinical and Experimental Radiologic Research, University of Vienna, Währinger Gürtel 18–20, A-1090 Vienna, Austria. Received August 14, 2003; revision requested October 28; revision received March 18, 2004; accepted May 17. Address correspondence to M.S. (e-mail: martina.scharitzer@meduniwien.ac.at).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
MATERIALS AND METHODS: The ethics committee did not require approval or patient informed consent when this study began. Because of a change in regulations, before images were acquired the nature of the study and procedures were explained to patients or their relatives, and consent was then obtained. A total of 131 catheter fragments (12–14 per radiograph) were superimposed over 10 anteroposterior bedside chest radiographs obtained with storage phosphor technology. Images were displayed on an LCD monitor (1536 x 2048 matrix) and a CRT monitor (2048 x 2560 matrix). Five radiologists independently located the catheter fragments and rated their confidence in detection with bright and subdued ambient light. A two-way analysis of variance and the Friedman test were used for statistical analysis.
RESULTS: There was no significant difference for either display type with respect to correctly detected catheter fragments (mean sensitivity, 56.6% and 56.0% for the CRT and the LCD monitors, respectively, with bright light and 61.2% for both monitors with subdued light). With both display types, detection rate with bright light decreased significantly (P < .05). False-positive rates and confidence ratings were not significantly affected by monitor type or ambient light.
CONCLUSION: In a study with simulation of clinical conditions, performance of the LCD monitor and high-resolution CRT monitor for detection of support catheters on bedside chest radiographs was equivalent. With both displays, detection performance was equally reduced with bright ambient light.
© RSNA, 2004
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
Active-matrix liquid crystal display (LCD) monitors are the newest technology for digital image display. Potential advantages of the active-matrix LCD over the CRT display are the elimination of peripheral distortion artifacts that may occur because of the curved surface of CRT monitors and reduced vulnerability to reflections and ambient light conditions. The most recent measurements suggest excellent spatial resolution and virtual elimination of veiling glare with LCD monitors (8). Yet, clinical experience with the new LCD monitor is limited.
Thus, the purpose of our study was to compare reader performance with an LCD monitor and with a high-resolution CRT monitor in the detection of simulated support catheters on bedside chest radiographs.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
|
For clinical evaluation, hard copies of the radiographs were printed. On these hard copies, the superimposed catheter fragments were specifically marked so as not to interfere with the diagnostic purpose of the image. All study images were of adequate quality to fulfill their diagnostic purpose. To further decrease the likelihood that the superimposed catheter fragments would obscure important clinical findings, we had selected study patients with stable cardiopulmonary function who underwent daily radiographic monitoring for other reasons.
Image Acquisition
The digital bedside chest radiographs were acquired by using storage phosphor technology (STVn; Fuji, Tokyo, Japan). Images had a matrix size of 4280 x 3520 pixels with 10 bits per pixel. The radiographs were obtained with a bedside unit (Mobilett II; Siemens, Erlangen, Germany), with a focus-film distance of 100 cm without automatic exposure control, 125 kVp, a tube current of 1.25–2.20 mAs, an antiscatter grid with 40 lines per centimeter, and a grid ratio of 10:1.
Images were processed by using the algorithm we routinely use for our bedside radiographs. The algorithm consists of a sigmoid gradation curve (gradation type, GT = E) and low-spatial-frequency enhancement (enhancement factor, RE = 0.3) with unsharp mask filtering and a kernel length of 2.5 cm (mask size, RN = 0).
Image Display
By using the picture archiving and communication system, the images were loaded on the following two monitor displays: (a) a high-resolution 5-megapixel gray-scale CRT monitor (HB 2183; Agfa, Mortsel, Belgium) with a matrix of 2048 x 2560 and a video card (MD2PCI2; Dome, Waltham, Mass) and (b) a 3-megapixel monochrome flat-panel display (C3TM; Dome) with active-matrix LCD technology, a matrix of 1536 x 2048, and a video card (DX/PCI; Dome).
For the sake of consistency of image appearance with the two display devices, we ensured that both display functions closely complied with the Digital Imaging and Communications in Medicine standard that was based on the Barten model (9). The maximum luminance was adjusted to 300 candela (cd)/m2 for both monitors, and the minimum luminance was set as low as possible and was close to 0.3 cd/m2. Maximum and minimum luminance values were set with the same ambient light conditions as those with which the reading process was conducted. The CRT faceplate was covered with an antireflective screen coating. The display size was equivalent for both monitors (30 x 40 cm).
The two monitors were located side by side on the same table so that light conditions were identical. No additional image processing was applied. No online processing, such as magnification or windowing, was available.
Reading Methods
The images were evaluated by five independent radiologists (including C.S.P., E.O., M.S.) with varying experience in chest radiography and digital imaging: two were faculty radiologists, with 15 and 6 years of experience in digital chest imaging, and three were residents, one 1st-year and two 3rd-year residents. To compensate for learning effects, the images were viewed in a different random order by each reader, with at least a 2-week interval between reading sessions. Reading was performed with subdued light with ambient light dimmed (luminance measured in front of the monitor, <20 lux) and with bright light with overhead lights switched on (luminance measured in front of the monitor, >100 lux). Each reader performed reading in two reading sessions, one with bright light and one with subdued light. In each reading session, images were evaluated in alternating order by using both monitor displays. Images were read in a random order, not pairwise, that was different for each reader. In each reading session, the reader saw 20 images, and viewing time per image was unlimited.
Three of the five readers saw the images with subdued light first, and the remaining two readers saw the images with bright light first. The readers were positioned at varying viewing distances according to their preferences and at a viewing angle of 90°.
The radiologists were asked to locate any support catheters and mark them on a transparent foil that was superimposed over the monitor display but could be removed easily and replaced in case of disturbing reflections. For each of the suspected detections, they were asked to rank their confidence on a three-point scoring scale as follows: 1, the catheter was probably present; 2, the catheter was present with high confidence; and 3, the catheter was definitely present. Prior to each reading session, readers saw three images in cases that were not part of the data analysis to become familiar with the reading procedure and to allow their eyes to adapt to the light conditions. The readers were familiar with the types of support catheters and knew only that fragments were superimposed. They did not know the number of fragments or the distribution of subtypes.
Statistical Analysis
The readers’ detection performance was assessed by means of the sensitivity and the false-positive rate on a reader-by-reader basis. The level of significance was set at a P value of .05. The sensitivity was defined as the percentage of correctly detected catheter fragments. The false-positive rate was defined as the percentage of false-positive ratings relative to all positive ratings. These numbers were calculated without consideration of the reader’s diagnostic confidence. For statistical analysis, two-way analysis of variance (ANOVA) was performed to assess the effect of ambient light and monitor type on the reader’s performance.
Significance of difference between the sensitivity values, the false-positive rates achieved with the four reading conditions (CRT display and subdued light, CRT display and bright light, LCD and subdued light, and LCD and bright light), and the effect of ambient light and monitor type on the level of the reader’s confidence were determined by using two-way repeated-measures ANOVA with a P value of less than .05.
Confidence ratings were averaged per reader, monitor type, and light condition; only true-positive ratings (scores 1–3) for detected catheter fragments were included in the analysis. Catheter fragments that had not been detected (false-negative ratings) were not considered.
The degree of interreader variability, with respect to the detection performance and the confidence ratings, was assessed by testing whether the distribution of ratings significantly differed for the five readers; the significance of the difference between these distributions was assessed by using the Pearson 
2 test, with a P value less than .05. Statistical analysis was performed with a statistical software package (SPSS Windows, version 11.5.1; SPSS, Chicago, Ill).
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
|
|
|
Confidence Levels
There was no significant difference between the mean confidence levels (2.48 and 2.49 for the LCD monitor and 2.50 and 2.53 for the CRT monitor, respectively, with the two light conditions). Results of two-way ANOVA revealed no significant effect of ambient light (P = .723) or monitor type (P = .122) on the readers’ confidence (Table 2). There was no significant interaction between monitor type and ambient light (P = .828).
Effect of Interreader Variability
There was no significant difference for the distribution of "yes" and "no" decisions among the five readers with respect to the presence of catheter fragments with the various reading conditions (P > .05, Pearson 2 test). However, the distribution of confidence ratings was significantly different for the five readers (P < .001, Pearson 2 test). This was true for all four reading conditions.
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
The most recently introduced active-matrix LCD monitors offer some ergonomic, financial, and display-related advantages compared with the traditional curved-surface CRT monitors, including higher luminance, a shorter depth of the monitor with a lower weight, and a flat-panel display (10). While researchers (11) in initial evaluations reported disadvantages, such as a lower signal-to-noise ratio, more veiling glare, and a lower uniformity, others (8) reported excellent performance with a superior spatial resolution (almost unity) and virtually no veiling glare.
Our results suggest an at least equivalent performance with a CRT monitor and an LCD monitor for the task of clinical detection of support catheters. This was true for the individual performance of the five readers, as well as for the averaged performance over all readers. Since results in previous reports already had suggested an equivalent overall diagnostic performance (10), we wanted to detect small performance differences by using a rather difficult detection task. We superimposed only support catheter tips because the correct detection of the tip is one of the most frequent diagnostic tasks in an intensive care unit. The relatively low detection rates (sensitivity values of approximately 60%) suggest that most of the fragments were difficult to detect. Reasons for the low sensitivity rates are that we used only low-density catheters that were not highlighted by an opacifying stripe. In addition, we used only the tips of support catheters and only relatively short fragments of approximately 2 cm in length. Both high- and low-attenuation areas of the thorax were equally covered.
Our results, which demonstrate equivalent detection performance for support catheter fragments, are in agreement with those of previous reports in which an equivalent detection performance for intrapulmonary nodules was described (12).
Veiling glare and ambient light reflection significantly degrade the display quality by increasing the luminance in black regions, with a consequently reduced contrast ratio of the monitor (difference between the maximum and minimum luminance values) and diminished structural display contrast. As a result, discrimination of low-contrast stimuli, such as catheter fragments, is reduced as ambient light is increased (13–15). A reduction of the contrast sensitivity of the human eye as a function of increasing background luminance further supports this fact (16). Accordingly, we found a substantially reduced detection rate with both monitors and bright ambient light.
As opposed to the CRT emissive technology, LCDs are light-modulating devices that form the image on the screen by controlling the transparency of the individual pixel and compensating for bright ambient light by absorbing the ambient light. In a previous study (16), researchers had found a superior discriminability of gray-scale differences with the LCD monitor compared with the CRT monitor with bright ambient light. The monitors tested in this previously published study were characterized by relatively low luminance values (100 cd/m2 for the CRT monitor vs 250 cd/m2 for the LCD monitor) and a rather low spatial resolution.
In our study, we could not confirm superiority of the LCD monitor over the CRT monitor with bright ambient light. This is most likely because we had set the maximum brightness of the LCD monitor at 300 cd/m2, and in that way, we potentially underestimated its ability to compensate for reflections of ambient light. We had set the maximum brightness to 300 cd/m2 for both displays in an attempt to ensure an identical contrast perception of the two monitors and to prevent eye fatigue caused by brightness that was too high with the LCD monitor. Results in previous studies had indicated that the luminance at which the display operates did not significantly influence the final diagnostic conclusion (17) but did influence reading time and speed of performance. We therefore conclude from our data that, with the parameter sets as determined in our study, both displays are optimized for a darkened diagnostic reading room, and different parameters have to be used for the LCD monitor with bright ambient light conditions to prove its advantages.
The significantly different distribution of confidence ratings indicates a high interreader variability and different patterns of reader behavior. Potential differences in reader confidence might have been obscured by the reader variability, and a larger number of readers would be required to prove a significant difference.
We found a higher number of false-positive rates with the LCD monitor for both ambient and subdued light conditions. ANOVA results, however, did not reveal a significant difference, which might be caused by the small number of readers and the considerable performance variability. A possible explanation for the higher false-positive rates with the LCD monitor is that the readers were less familiar with that monitor. Accordingly, all readers reported a subjectively higher familiarity with the CRT monitor compared with the LCD monitor. Other factors such as off-angle viewing, a larger viewing distance, or different light conditions had been excluded.
Possible distortion in curved-surface displays was not subjectively evaluated and did not appear to significantly affect detection performance. However, it should be noted that we did not specifically address this problem and that other clinical indications, such as stereotactic biopsy or orthopedic applications, may be more susceptible to these types of artifacts than chest radiography (11).
The limitations of our study include the following issues. First, that we included five readers and 10 posteroanterior radiographs was a limitation. We tried to offset this limitation by including readers who had varying experience. We also attempted to offset the limitation by using multiple fragments of different types of support catheters and locating them in diverse positions for the posteroanterior radiographs to simulate a variety of conditions. Since the study setup focused on the detection of only catheter tips and did not involve a complex diagnostic task, we considered these procedures appropriate.
Second, our results suggest an equivalent performance of the two types of monitors, although it has to be noted that our results are valid only for a comparable setup in which the two monitors, LCD and CRT, have matched gradation characteristics.
Third, potential preference bias could not be completely eliminated, since both types of monitors could be readily identified.
Fourth, we tested only a very specific diagnostic task, and further studies are needed that would include other diagnostic or interpretive challenges.
Last, although online processing, such as magnification or windowing, represents an essential part of soft-copy reading, it was not included in our study setup. This was done in an attempt to facilitate the study design. Use of online windowing would have most likely decreased the effect of ambient light but also would have introduced other factors, such as individual reading habits and variable familiarity with the use of the workstation, that have an effect on detection performance but are difficult to control. Furthermore, performance differences or differences in confidence between the two monitors are likely to decrease when online windowing is available. Also the false-positive rate is likely to be lower with both types of monitors when online windowing is available, a fact that would have been true for detection with both types of monitors.
Reading sessions were rather short (<45 minutes), and other aspects, such as eyestrain and overall fatigue, which have been reported (11) to be greater for the CRT display, did not play a role in our study. Another aspect worth considering is the effect of off-angle viewing. The flat-panel display is more dependent on a rectangular viewing angle to achieve optimal performance compared with the curved-surface monitor. Viewed straight on, on-axis contrast ratios can be as high as 300:1, but if the viewer moves off axis, the contrast significantly drops. In the study situation, the effect of off-angle viewing was effectively eliminated by correct placement of the reader directly on axis in front of the monitor; the placement potentially becomes more important in a busy radiology department.
In conclusion, in a study with simulation of clinical conditions, performance of both the flat-surface LCD monitor and the traditional curved-surface CRT monitor for detection of support catheters on bedside chest radiographs was equivalent. Both monitor displays had an equal decrease in detection performance with bright ambient light when luminance values had been optimized for dark ambient light. The readers’ diagnostic confidence appeared to be equal for displays of both monitors and was independent of ambient light conditions.
|
FOOTNOTES |
|---|
Authors stated no financial relationship to disclose.
Author contributions: Guarantors of integrity of entire study, M.S., C.S.P.; study concepts and design, C.S.P.; literature research, M.S., E.O.; experimental studies, M.S., C.S.P., M.P.; data acquisition, M.S., E.O., M.F.; data analysis/interpretation, M.W., M.P.; statistical analysis, M.W., M.P.; manuscript preparation and definition of intellectual content, M.S., C.S.P.; manuscript editing, C.S.P.; manuscript revision/review, E.O., M.F., M.P., M.W.; manuscript final version approval, all authors
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  N Buls, W Shabana, P Verbeek, P Pevenage, and J De Mey Influence of display quality on radiologists' performance in the detection of lung nodules on radiographs Br. J. Radiol., September 1, 2007; 80(957): 738 - 743. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 P. C. Brennan, M. McEntee, M. Evanoff, P. Phillips, W. T. O'Connor, and D. J. Manning Ambient Lighting: Effect of Illumination on Soft-Copy Viewing of Radiographs of the Wrist Am. J. Roentgenol., February 1, 2007; 188(2): W177 - W180. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 H. P. McAdams, E. Samei, J. Dobbins III, G. D. Tourassi, and C. E. Ravin Recent Advances in Chest Radiography Radiology, December 1, 2006; 241(3): 663 - 683. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
