| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Original Research |
1 School of Medicine and Medical Sciences, University College Dublin, Health
Science Bldg., Belfield, Dublin 4, Ireland.
2 American Board of Radiology, Tucson, AZ 85711.
3 School of Medical Imaging Sciences, St. Martin's College, Lancaster, UK LA1
3JD.
4 Applied Neurotherapeutics Research Group, Conway Institute, University College
Dublin, Dublin 4, Ireland.
Received November 23, 2005;
accepted after revision March 15, 2006.
Address correspondence to P. C. Brennan.
Abstract
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
MATERIALS AND METHODS. Five ambient light levels were investigated: 480, 100, 40, 25, and 7 lux. Seventy-nine experienced radiologists were asked to examine 30 posteroanterior wrist images and decide whether a fracture was present. All images were displayed on liquid crystal display monitors. Receiver operating characteristic analysis was performed, and the numbers of false-positive and false-negative findings were recorded.
RESULTS. For all the radiologists, greater area under the receiver operating characteristic curve and lower numbers of false-positive and false-negative findings were recorded at 40 and 25 lux compared with 480 and 100 lux. At 7 lux, the results were generally similar to those at 480 and 100 lux. The experience and knowledge of radiologists specializing in imaging of musculoskeletal trauma appeared to compensate in part for inappropriate lighting levels.
CONCLUSION. Typical office lighting and current recommendations on ambient lighting can reduce diagnostic efficacy compared with lower levels of ambient lighting. If, however, no light other than that of the monitor is used, results are similar to those with excessive levels of lighting. Careful control of ambient lighting is therefore required to ensure that diagnostic accuracy is maximized, particularly for clinicians not expert in interpreting posteroanterior wrist images.
Keywords: lighting • PACS • wrist • X-ray technology
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The effect of high levels of ambient lighting on diagnostic efficacy has been well reported. Illumination of 450 lux reduces visualization of micronodules and pulmonary lines of chest images on cathode ray tubes [3]. Detection of catheters with images on cathode ray tubes is reduced with bright compared with subdued ambient lighting [4], and detection of nodules is significantly decreased with higher levels of ambient lighting [5]. The need for adherence to suitably low ambient lighting is clear.
Two key standards appear to be in place for minimizing the effect of ambient lighting on interpretation of general radiologic images: the World Health Organization [6] states that ambient lighting should be below 100 lux 30 cm from the image, and the Commission of the European Communities [7] requires a level below 50 lux 100 cm from the display. These standards are useful and have been put to good effect in previous work [8, 9]; however, the standards were set at a time when film was the predominant medium for displaying images. Film is traditionally viewed on a transilluminator with a luminance of up to 3,000 cd/m2. In an environment of softcopy reporting, in which the luminance of cathode ray tubes and liquid crystal displays is much lower than that of a transilluminator, the relevance of the current standards is unclear. In addition to the aforementioned standards, the standard of the American College of Radiology [10] is that for analog mammograms ambient lighting should be 50 lux with a display luminance of 3,000 cd/m2 (at the viewbox surface). The standard for digital mammograms is that lighting should be at a low level and diffuse. No specific values are given for digital images.
Our hypothesis was that ambient lighting in which general radiologic images are viewed and clinical decisions are made is inappropriate and lacks effective standards. The aim of this study was to examine the relevance of existing standards for visualization of wrist fractures and to propose more relevant values if these are required.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Image Display Conditions
All images were displayed on one of three liquid crystal display monitors
(VG810b, Viewsonic) with a resolution of 1,280 x 1,024 pixels and a
refresh rate of 72 Hz. At each ambient light setting, the monitors were
calibrated to the DICOM grayscale display function standard
[11] with Verilum software
(Image Smiths) and luminance pod. Quality checks performed throughout the work
with the Society of Motion Picture and Television Engineers test pattern
[12,
13], American Association of
Physicists in Medicine Task Group 18
[14] findings, and a
calibrated photometer (model 07-621, Nuclear Associates) showed for each
monitor that maximum luminance did not fall below 250 cd/m2;
minimum luminance did not exceed 1 cd/m2; and geometry, luminance
uniformity, temporal stability, resolution, and veiling glare remained within
recommended levels [14]. The
rooms in which all measurements were made were painted with a light yellow
matte finish and had an area of approximately 16 m2.
Ambient Lighting
Five levels of ambient lighting were investigated: 480 lux, normal office
lighting; 100 lux, current recommendation
[6]; 40 lux; 25 lux; and 7 lux,
absence of any lighting apart from the monitor. Forty lux and 25 lux are
typical values in the darkened rooms used by consultant radiologists primarily
for image reporting. Ambient light levels were achieved with artificial
lighting controlled by a power strip with surge suppression and elimination of
any external and natural light from the study room. The light arrangement did
not result in direct glare on the monitor. Illumination was monitored with the
calibrated photometer. A wrist image was displayed on the monitor, and all
measurements were within 10% of the stated values. Measurements were taken 30
cm from the display with the long axis of the photometer aligned and
perpendicular to the center of the monitor.
Image Evaluation
A test bank of 30 posteroanterior wrist images was assembled from a
collection of 740 cases of radiographs of the lower forearm collected from the
archive of a single district general hospital. These hard-copy images were
divided into normal (no fracture present) and abnormal (fracture of lower end
of radius, ulna, or both) groups. The diagnostic groups were defined initially
on the basis of the radiology reports on the cases. The reports were verified
by follow-up record from the file on clinical treatment in each case. The
ground truth for the diagnosis was determined, therefore, by both radiologic
opinion and clinical outcome.
The test bank was converted to a digitized format in the following way: Each hard-copy image was recorded from a standard light box in controlled ambient lighting conditions to a charge-coupled device monochrome camera (Hitachi Denshi) with a zoom lens (18-108/2.5, Sony). This camera was connected to a Sprynt imaging framestore (Synoptics) operating within the program, PC-Image 2.1 (Program4Pc), running on a standard PC with a Pentium 4 processor (Intel). Each image was digitized in 16-bit form and stored as an integration of 15 acquisition frames to minimize additional noise and was saved uncompressed in the TIFF format. No subsequent contrast optimization or other image processing was performed. Cropping to standard dimensions resulted in an average file size of 180 kB per image (range, 157-294 kB).
Thirty of these images were presented in a random order to each of the viewers in the experiment. Fifteen images showed wrist fractures of varying subtlety, and 15 images showed a normal wrist. All fractures were visible on the posteroanterior image. The prevalence of fracture in the test bank was not revealed to the observers. Technical criteria of correct positioning and exposure of the patient and inclusion of all the carpal bones and the distal third of the radius and ulna were applied to all images before they were included in the study. Each image was masked so that any white display around the wrist was removed.
Seventy-nine radiologists who had been certified by the American Board of
Radiology for a mean of 23.1 years (minimum, 9 years; maximum, 46 years), all
of whom were oral examiners for the board, assessed the images. A minimum of
15 radiologists were randomly allocated to a specific light level; at least
30% of these radiologists were specialists in examining musculoskeletal images
(Table 1) and were described as
musculoskeletal radiologists. Each radiologist assessed images at only one
light level to eliminate bias introduced by recognition of images. No
manipulation was used, and viewers were allowed to judge the images at any
distance from the monitor they found comfortable. With the use of software
developed at Lancaster University, each image was presented for a maximum of
30 seconds (considered sufficient time by all viewers), and a score of 1-10
was awarded (10 = fracture definitely present, 1 = fracture definitely not
present). Any score 
6 represented presence of a fracture. Once an image
was scored, the next image was displayed. Each session of analyses ended after
all images in the sequence had been shown, and a data file compatible with
Rockit (University of Chicago) was automatically generated for immediate
receiver operating characteristic analysis. The software is described at
polo.lancs.ac.uk/phillips/feedback/WristViewer/instructions.html.
Area under the receiver operating characteristic curve (Az) was
calculated, as were false-positive and false-negative rates for each
observer.
|
Statistical Analysis
One-way analysis of variance was used to compare Az,
false-positive, and false-negative values between pairs of ambient light
settings. Analysis was performed first for all radiologists involved in the
work and second for the musculoskeletal radiologists. A significance level of
p 
0.05 was set for all comparisons.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
0.024) (Table 2). No
statistical differences were found for the musculoskeletal radiologists.
Statistically higher numbers of false-positive findings were found at 480,
100, and 7 lux compared with 40 and 25 lux when all the radiologists were
compared (p 0.0001) (Table
3). No differences were found for the scores of musculoskeletal
radiologists only. No significant differences were found between ambient
lighting levels for either group of radiologists
(Table 4).
|
|
|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The results for the entire radiologist group show the need for correct ambient light levels. They also show that relying on existing standards available for film-based viewing of general radiologic images is not appropriate. A clear pattern emerged in which at 480 and 100 lux, Az values are reduced and numbers of false-positive and false-negative findings are increased compared with the findings at 40 and 25 lux. No benefit for this group of viewers was found at the recommended ambient light level over typical office room lighting. Our findings reasonably agree with values stated for specific specular and diffuse reflections in the American Association of Physicists in Medicine [14] report on monitors with luminance such as those described in this work. It can be argued that improvement in Az values at 40 and 25 lux compared with the brighter environment was subtle, only approximately 3% improvement in scores. If, however, one examines the false-positive data (fractures reported when none were present and the category in which most "mistakes" fall), 30-60% reduction in the false-positive rate was evident when lower light levels were used. This finding has important implications for image viewing practice in clinical centers. It appears that as many as 50% of all radiologic images can be examined in the nonradiologic office environment by non-musculoskeletal specialists and that clinical decisions can be made on the basis of findings at this examination.
The results at 7 lux (no ambient light apart from that provided by the monitor) produced intriguing findings. Although one might have expected the improvement in diagnostic efficacy at 40 and 25 lux over 480 and 100 lux to continue or at least level off at 7 lux, this phenomenon did not occur, apart from false-positive findings by the specialist group. Apart from the exception noted, the values gained at this very low level of lighting were similar to those at 480 and 100 lux. Although the reason for reduced efficacy at this low illumination is not clear from the current study and further work is required, this result may be linked to the eye fatigue often experienced with prolonged watching of television without ambient lighting or stargazing on a dark night. It should be emphasized that in addition to reduced diagnostic accuracy, there are health and safety considerations of working with this low lighting.
Our findings were statistically significant only for all radiologists and not for the specialist group, although overall trends were not dissimilar for both groups. Examination of the nonsignificant findings for the musculoskeletal specialist group shows, however, that the differences in Az values between light levels are less obvious, particularly for all illuminations in the range of 7-100 lux. Although nonsignificant differences in values are evident, it appears that the experience and knowledge of the musculoskeletal specialists compensate at least in part for inappropriate light levels. The relevance of these results to nonspecialist and musculoskeletal radiologists making clinical decisions in inappropriate lighting conditions is clear.
Az values and false-positive and falsenegative results did not show as close agreement as would be expected in a comparison of the two groups of radiologists. For example, at 40 lux there was little difference between the two groups for Az values, but false-positive rates were quite different. The reason is unclear, but it is probably linked to the way the musculoskeletal radiologists were more likely to use the extremes of the response scale—that is, scoring 1 and 10—because they had greater confidence viewing these types of images than did the other radiologists. The nonspecialist radiologists used midrange values to a larger extent than the specialists. This finding shows the importance of looking at several parameters to measure observer performance.
Although our results provide guidance for ambient light levels for wrist examinations, which account for approximately 15 million examinations in the United States each year, the usefulness of the recommendations for viewing of other images, such as those of the breast, is not described. On the basis of the current results, we anticipate that the trends found are likely to be relevant in other situations, but more research is needed for establishment of examination-specific guidance values. In addition, our results are relevant only to the consumer-grade monitors described in the methods. Professional-grade monitors may require alternative recommendations because of different spatial, contrast, and surface characteristics.
In summary, this study showed that for optimum viewing of radiologic images of the wrist to identify fractures, ambient lighting of 25-40 lux should be used. This level of lighting is comfortable and facilitates the multitasking environment of radiologic reporting.
Acknowledgments
We are very grateful to the trustees, oral examiners, and staff of the
American Board of Radiology, who facilitated the work. Thanks are due to all
the radiologists who contributed their time to interpret the images and
participate in the work so enthusiastically.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
This article has been cited by other articles:
|
|
 |
|
  M. Bath EVALUATING IMAGING SYSTEMS: PRACTICAL APPLICATIONS Radiat Prot Dosimetry, February 10, 2010; (2010) ncq007v1. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
D. P. Chakraborty CLINICAL RELEVANCE OF THE ROC AND FREE-RESPONSE PARADIGMS FOR COMPARING IMAGING SYSTEM EFFICACIES Radiat Prot Dosimetry, February 5, 2010; (2010) ncq017v1. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
D. P. Chakraborty A STATUS REPORT ON FREE-RESPONSE ANALYSIS Radiat Prot Dosimetry, January 18, 2010; (2010) ncp305v1. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
