and (4) Latitude CP Laptop (Dell Computer Corp.). Raters were allowed to magnify and to adjust density and contrast of each image at will. Receiver Operating ...
Dentomaxillofacial Radiology (1999) 28, 203 ± 207 ã 1999 Stockton Press All rights reserved 0250 ± 832X/99 $12.00 http://www.stockton-press.co.uk/dmfr
In¯uence of the digital image display monitor on observer performance RA Cederberg*,1, NL Frederiksen1, BW Benson1 and JD Shulman1 1
Baylor College of Dentistry/Texas A&M University System, Dallas, Texas, USA
Objective: To assess the in¯uence of the display monitor on observer performance. Material and methods: Arti®cial enamel lesions were created in 40 extracted teeth at random using and round burs. Teeth were mounted in dental stone blocks to simulate a hemidentition. Approximate exposures were recorded at 70 kVp using a Digota (Soredex, Orion Corp, Helsinki, Finland) digital imaging system, calibrated to achieve optimum density. Six dentists rated each image on a ®ve-point scale for the presence or absence of a lesion. Radiographic images were viewed on the following monitors: (1) AlphaScan 711 (Sampo Corp.); (2) Multiscan 17 Se II (Sony Electronics Inc.); (3) DS 2000 (Clinton Electronics Corp.) and (4) Latitude CP Laptop (Dell Computer Corp.). Raters were allowed to magnify and to adjust density and contrast of each image at will. Receiver Operating Characteristic (ROC) analysis was performed and curves were plotted for each image. Data was subjected to repeated measures analysis of variance and ordinal logistic regression to test for signi®cance between variables and to determine odds ratios. Results: Mean ROC curve areas ranged from 0.8728 for the Sampo monitor to 0.8395 for the Sony. Repeated measures analysis of variance showed signi®cant dierences between observers (P50.0001), lesion size (P50.0001), examiner/monitor interaction (P50.033) and examiner/ block interaction (P50.013). However, no signi®cant dierence was found between monitors. Conclusion: This study suggests that observer performance is independent of the visual characteristics of the display monitor. Keywords: digital radiography, dental; ROC curve; diagnostic imaging; perception
Introduction Direct digital imaging continues to gain acceptance in dentistry. However, all the commercial systems currently available have as a limitation inferior spatial resolution compared with radiographic ®lm. The spatial resolution has been reported to vary from 6 to 10 line pairs per millimeter (lp/mm) depending on the system.1 Whereas ®lm is up to 20 lp/mm.2 The eect of resolution on observer performance is equivocal. In a recent study that used simulated enamel lesions, it was reported that ®lm out performed a photostimulable phosphor (PSP) digital system.3
*Correspondence to: RA Cederberg, Baylor College of Dentistry/Texas A&M University System, P.O. Box 660677, Dallas, Texas. TX 75266-0677, USA Received 28 December 1998; accepted 3 March 1999
The video display used for Picture Archiving and Communication Systems (PACS) in the medical radiography ®eld has been reported to be the weak link of the system.4 However, no study has reported on the eect of the display monitor with direct digital images in dentistry. Factors which determine monitor ®delity, i.e. accuracy of reproduction, include, but are not limited to: monitor resolution (number of vertical lines, bandwidth and refresh rate), bit depth, dot pitch, luminance, and display size. In addition to monitor ®delity, observer performances may also be limited by the resolving power of the human visual system. It has been reported that the minimum contrast threshold of the human eye corresponds to a spatial frequency of 5 lp/cm (a 1 mm wide line paired with a 1 mm wide space) which would correlate to a monitor pixel size of about 1 mm. Most current cathode ray tube (CRT) displays used with dental direct digital imaging systems have a pixel size of 0.3 mm, and the eye has been
Display monitors RA Cederberg et al
204
reported to become less sensitive to pixel sizes smaller or larger than 1 mm.5 Film resolution is limited to the inherent resolution of the ®lm or screen-®lm combination. For digital systems, resolution is determined by both image detector and monitor resolution.5 Most commercially available monitors have resolutions (pixel matrix size) of 10246768, but some high performance monitors are available that have pixel matrices as high as 204862048.6 These high resolution monitors are relatively expensive and are unlikely to be sold to practicing dentists. Luminance refers to the amount of visible light emitted by the CRT display and is expressed in candela per square meter (cd/m2). Most gray-scale monitors commonly used with dental digital imaging systems have a maximal luminance that ranges from 86 cd/m2 to 240 cd/m2, compared with 1542 cd/m2 to 1713 cd/m2 for typical ®lm viewboxes.6 Luminance aects the observer's response through the acuity (ability to detect ®ne details) of the eye. Up to a point, the greater the monitor's background luminance, the less sensitive the eye is to absolute or just noticeable dierences. Just noticeable dierences increase linearly with luminance. However, the image is perceived by the eye to be `better' in quality up to approximately 3500 cd/m2, then declines possibly due to excessive glare.5 Thus, background luminance levels may in¯uence observer's ability to distinguish ®ne details and just noticeable dierences on the display screen.6 To the contrary, one study has reported a slight drop in detection rates of test images above 274 cd/m.2,7 Comparison of image quality of CRT displays and ®lm is most often achieved with the use of two psychophysical techniques either clinical images and ROC analysis or contrast-detail patterns to determine threshold contrast.6 In our previous study3 arti®cial lesions were chosen as the test phantom because wellde®ned, high contrast lesions allowed observers to discriminate between sound enamel and a cavitated surface. For this reason, we have used the same approach again. The objective of the present study was to assess the in¯uence of the display monitor on observer performance. Materials and methods Arti®cial lesions con®ned to enamel were created on 27 approximal surfaces of 40 caries-free (as determined by visual inspection) extracted human teeth using either a round (0.5 mm diameter) or a round (0.2 mm in diameter) bur and sinking the head of the bur to the junction of the neck. Teeth were mounted in groups of
eight in ®ve stone blocks to simulate a hemidentition. The number of lesions per block varied from ®ve to six. Numbers of or round lesions were randomly distributed in each block. Each block was imaged using a PSP imaging plate (Digora, Orion Corporation Soredex, Helsinki, Finland). The X-ray source was a IRIX 70-E operated at 70 kVp and 8 mA (Trophy Radiologie, Vincennes Cidex, France). Exposure techniques were adjusted to achieve an optical density of 1.0 in dentin. Four sheets of 3 mm thick masonry board were placed between the object and the focal spot of the X-ray tube to act as a scattering medium.8 Each of the ®ve blocks was imaged with the long axes of the teeth parallel to the imaging plate and the central ray of the X-ray beam directed perpendicular to the plane of the plate. A total of six dentists (two oral pathologists, two oral radiologists and two general practitioners) scored each of the ®ve images displayed on four dierent monitors: (1) AlphaScan 711 (Sampo Corporation, Taipei, Taiwan); (2) Multiscan 17 Se II (Sony Electronics Inc., Park Ridge, NJ, USA); (3) DS 2000 (Clinton Electronics Corporation, Rockford, IL, USA) and (4) Latitude CP Laptop (Dell Computer Corporation, Round Rock, TX, USA). The speci®cations of the monitors are shown in Table 1. Each image was scored for the presence or absence of approximal lesions on a ®ve-point scale: 1=de®nitely present; 2=probably present; 3=cannot tell; 4=probably not present; 5=de®nitely not present. Observers were instructed to adjust density, contrast and magni®cation of digital images to their preference. All images were viewed using the unenhanced mode of the Digora software display. The mean score for the ®ve blocks was used to construct ROC curves (ROCFIT, Charles E. Metz, Dept. of Radiology, University of Chicago) for each monitor type. Data was subjected to repeated measures analysis of variance (ANOVA) to test for dierences between observers, lesion size, examiner/monitor interaction and examiner/block interaction. Ordinal logistic regression9 was used to estimate the extent to which the monitor aected the odds of detecting and round lesions. The absolute value of the dierence between observer score (1 ± 5) and whether a lesion was present (1) or absent (5) on each surface was used as the outcome (dependent) variable. Higher scores indicated greater degrees of disagreement (i.e., error) on the part of the observer. Table 2 illustrates the ®ve-level (zero to four) coding scheme. For example, both an observer who scored a surface with a lesion as `4' (probably not present) and one who scored a surface without a lesion as `2' (probably a
Table 1 Specifications of the four CRT Monitors Type
Screen size
Sampo AlphaScan 711 Sony multiscan 17 Se II Clinton DS 2000 Dell latitude CP laptop
17 inch 17 inch 17 inch 13.5 inch active matrix
Resolution 10246768 8006600 10246768 10246768
(16 (16 (16 (16
Dot pitch bit) bit) bit) bit)
0.25 0.25 0.26 0.26
Video card size 4 2 4 2
mb mb mb mb
Luminance 274 103 274 72
cd/m2 cd/m2 cd/m2 cd/m2
Display monitors RA Cederberg et al
lesion present) would receive the same score 4 ± 5 and 1 ± 2 , respectively. Odds ratios were calculated for dierences in monitor, lesion size and lesion size/ monitor interaction. Results The mean Roc curve areas (AZ) for the four monitors were as follows: Sampo (0.8728), Laptop (0.8532), Clinton (0.8448) and Sony (0.8395) Figure 1. There was no signi®cant dierence between them. When subjected to repeated measures analysis of variance, signi®cant dierences were seen between observers (P50.0001), lesion size (P50.0001), examiner/monitor interaction (P50.033), and examiner/ block interaction (P50.013). Since no dierence was
Table 2 Coding scheme for ordinal logistic model Observer score 5 4 3 2 1
No lesion (5) Absolute Score-lesion value 5 ± 5=0 4 ± 5=71 3 ± 5=72 2 ± 5=73 1 ± 5=74
0 1 2 3 4
Lesion (1) Absolute Score-lesion value 5 ± 1=4 4 ± 1=3 3 ± 1=2 2 ± 1=1 1 ± 1=0
4 3 2 1 0
Figure 1 ROC curves for each monitor tested. ROC curve areas (AZ) are Sampo 0.8728, Laptop 0.8532, Clinton 0.8448 and Sony 0.8395
found between monitors, lesion size/monitor interaction was not included in the model. The results of the ordinal logistic regression to estimate odds ratios for lesion size and monitor are shown in Table 3. Observers were one-third as likely (OR=0.334) to detect quarter-round lesions compared with detecting no lesion or half-round lesions; less than one-half as likely (OR=0.474) to detect a half-round lesion compared with detecting no lesion; and approximately 70% more likely to detect half-round than a quarter-round lesion. The odds ratios for the monitors were not statistically dierent from unity, the value under the null hypothesis. Discussion The display and manipulation of digital images on a computer monitor is an essential element of computed radiography. The quality of an image is a function of the physical parameters of the system. The image sensor, computer hardware and software and external factors such as extraneous light and screen re¯ection, as well as the inherent limitations of the human visual system, all in¯uence image quality. This study was designed to determine whether monitors with a relatively superior ®delity could enhance observer performance. The four monitors tested in this study were chosen as a cross section of those commercially available monitors that would likely be used with dental direct digital imaging systems. The Sampo was representative of the average monitor that might come with many computer systems. While the Sony is typical of a relatively more expensive option. The Clinton monitor was selected because it is a high quality black and white monitor in widespread use in medical radiology. Lastly, the Dell laptop was included because some digital systems are now being marketed for use with portable laptop computers. Spatial resolution, screen size, bit depth, dot pitch and luminance are characteristics of monitors which may aect image quality. The spatial resolution of a monitor is most often expressed in terms of the size of the pixel matrix. High performance monitors are available with pixel matrices as high as 204862048. However, the resolvable pixel matrix of such monitors is considerably smaller.6 Commonly used monitors available for most computer systems have a nominal resolution ranging from 6406480 to 160061200. Size
Table 3 Results for the ordinal logistic regression Parameter
Odds ratio
95% Confidence interval Lower limit Upper limit
Wald Chi-square
Pr4Chi-square
a
Monitor Laptop Sampo Clinton Burb Half Quarter a
0.303 0.2032 70.0189
1.031 1.225 0.981
0.774 0.917 0.738
1.372 1.637 1.305
0.0431 1.8873 0.0168
0.8388 0.1695 0.8969
70.7469 71.0957
0.474 0.334
0.37 0.258
0.608 0.433
34.5845 69.3042
0.0001 0.0001
Compared with Sony monitor; bcompared with no lesion
205
Display monitors RA Cederberg et al
206
of the monitor screen as well as type and memory size of the video card de®ne or limit a monitor's resolution. All commercially available monitors are limited to a display of 256 gray shades. Monitors with a bit depth of 16 will be capable of displaying 65 000 colors, but for radiographic images, gray levels are limited to 256. Wenzel10 found that indirectly acquired digital images displayed at a spatial resolution of 5126512 and 64 shades of gray was equal to, or in some cases more accurate than, the original radiograph for the detection of bone lesions. The four monitors used in this study had almost the same resolution and gray scale display which was adequate for valid diagnosis and, as expected, did not in¯uence the results of this study. Dot pitch is a factor, that has an in¯uence on image quality, but only when the dierence between two monitors is signi®cant. Dierence of 0.01 for the monitors in this study is probably not perceptible considering the diagnostic task in question. Luminance has been a topic of much discussion and research in the medical imaging arena. The consensus up until the early 1990s was that a medical display of up to 514 cd/m2 was needed to match the viewing conditions of a conventional ®lm view box. Consequently, computer workstations for use in medical radiology have sought to develop CRT displays with a high maximum luminance. There are some monitors available for these PACS workstations that have maximum luminance in the range of 500 to 600 cd/ m2; however these monitors are expensive.5 All CRT displays should be capable of generating linear luminance from 0 to the maximum luminance of the display. However, phosphors used in these display devices are highly nonlinear and require electronics or software correction. The maximum luminance of the Clinton and the Sampo was 274 cd/m2. They were used for comparison in this study, because they have maximum luminance in the upper range of commonly used monitors. Theoretically, a maximum luminance dierence of 202 cd/m2 (274 cd/m2 for the Clinton and Sampo and 72 cd/m2 for the laptop) should make a perceptible dierence in the ability to detect ®ne detail in the image, but none was found. Modulation transfer function (MTF) measures the combined eects of sharpness and resolution. In practical terms, the ability of the display monitor to reproduce ®ne detail in an image depends less on the luminance it produces than on the MTF, ®delity with which the signal is recorded and the level of image noise. Brettle et al.11 found that PSP systems were
limited to a resolution between 6.3 and 7.1 lp/mm. The MTF of E-speed ®lm, is superior to that of PSP systems at high spatial frequencies, with the 50% modulation level for latter being reached at 2.5 cycles mm compared with well beyond 10 lp/mm for the former.11 These inherent limitations of PSP systems, as well as the type and size of the simulated lesions evaluated in this study, may oset the perceived advantages oered by superior ®delity of some CRT displays. It has been reported that larger simulated lesions are easier to detect than smaller ones.3 Likewise, this study found that observers were 70% more likely to detect round lesions compared with round lesions. This dierence in detection rates may be attributable in part to the dierences in contrast seen with simulated lesions compared with natural caries. A recent study12 found that with the Digora system there was little dierence between the detection rates of arti®cially and natural lesions in unenhanced images. In addition, this study con®rmed our ®ndings that regardless of whether the lesion was arti®cial or natural, deeper lesions were easier to detect. A display monitor study of this type should be repeated with natural lesions. In the medical imaging ®eld, the trend is toward the development of workstation CRT displays with improved image parameters such as luminance, display function, dynamic range, distortion, resolution and noise. These high ®delity monitors have made a dierence in the detection rates in mammography and chest images.4 In this study, the type and ®delity of the monitor used with a direct digital imaging system did not alter observers' diagnostic ability. A wide variety of choices are available to the practitioner when deciding on computer hardware for the dental oce. When choosing a CRT display for the computer system, the better the parameters of ®delity, the higher the cost. For example, while the average cost of computer monitors is about $500, the Clinton monitor tested in this study will cost about ®ve times that amount. If a direct digital imaging system is included in the diagnostic armamentarium, choice of a CRT display should not be a limiting factor.
Acknowledgements The authors wish to thank Soredex Inc., Conroe, Texas, USA for use of the Digora Imaging System, as well as two of the monitors tested in this study.
References 1. Wenzel A, GroÈndahl H-G. Direct digital radiography in the dental oce. Int Dent J 1995; 45: 27 ± 34. 2. Goaz PW, White SC. Oral Radiology: Principles and Interpretation 3rd edn. St. Louis Mosby: 1994, p 273. 3. Cederberg RA, Frederiksen NL, Benson BW, Shulman JD. Eect of dierent background lighting conditions on diagnostic performance of digital and ®lm images. Dentomaxillofac Radiol 1998; 27: 293 ± 297. 4. Arenson RL, Chakraborty DP, Seshadri SB, Kundel HL. The digital imaging workstation. Radiol 1990; 176: 303 ± 315.
5. Wang J, Langer S. A brief review of human perception factors in digital displays for picture archiving and communications systems. Digital Imag 1997; 10: 158 ± 168. 6. Dwyer III SJ, Stewart BK, William MB. Performance characteristics and image ®delity of gray-scale monitors. Radiographics 1992; 12: 765 ± 772. 7. Nishikawa K, Kuroyanagi K. Luminance dependency of detectability of CRT display. In: Farman AG, Ruprecht A, Gibbs SJ and Scarfe WC (eds). IADMFR/CMI'97 Advances in Maxillofacial Imaging Amsterdam: Elsevier 1997; 293 ± 298.
Display monitors RA Cederberg et al
8. Cederberg RA. Temporomandibular joint space analysis. J Crainomandib Pract 1994; 12: 172 ± 177. 9. Hosmer D, Lemeshow S. Applied logistic regression. New York, Wiley: 1989. 10. Wenzel A. Eect of varying gray-scale resolution for detectabilty of bone lesions in intraoral radiographs digitized for teletransmission. Scan J Dent Res 1987; 95: 483 ± 492.
11. Brettle DS, Workman A, Ellwood RP, Launders JH, Horner K, Davies RM. The imaging performance of a storage phosphor system for dental radiography. Br J Radiol 1996; 69: 256 ± 261. 12. Kang BC, Goldsmith LJ, Farman AG. Observer dierentiation of mechanical defects versus natural dental caries cavitations on monitor-displayed images with imaging plate readout. Oral Surg Oral Med Oral Pathol Oral Radiol Endod 1998; 86: 595 ± 600.
207