Dentomaxillofacial Radiology (2003) 32, 242–246 q 2003 The British Institute of Radiology http://dmfr.birjournals.org
RESEARCH
Influence of displayed image size on radiographic detection of approximal caries R Haak*,1, MJ Wicht1, G Nowak1 and M Hellmich2 1 2
Centre of Dental Medicine, Department of Operative Dentistry and Periodontology, University of Cologne, Ko¨ln, Germany; Department of Medical Statistics, Informatics and Epidemiology, University of Cologne, Germany
Objectives: To evaluate the validity of approximal caries detection on digital bitewing radiographs displayed at different image sizes on either a cathode ray tube (CRT) monitor or a thin film transistor (TFT) monitor. Methods: Five observers assessed digital radiographs of a charge-coupled device (CCD)-based sensor system (Sidexisw) of 160 unrestored premolars and molars for approximal caries using a six category caries rating scale. Images were displayed at ratios of 1:1, 1:2 and 1:7 on a CRT monitor (Nokia 446 XS) and a TFT display (Panasonic LC 50S). Histological assessments of serial sections were used as the validation standard. Diagnostic accuracy was expressed as area under the receiver operating characteristic (ROC) curve (AUC) and was calculated at two levels of caries penetration: presence of caries (I) and presence of a lesion in the dentine (II). The influence of the factors “monitor type”, “image size” and “validation threshold” were analysed with repeated measures analysis of variance. Results: The ROC curve areas for approximal caries detection at both histological penetration levels were not influenced by the type of monitor display, whereas image size had a significant impact (P , 0.01). AUCs for image size 1:7 (I, 0.62; II, 0.65) were smaller compared with ratios of 1:1 and 1:2 (P , 0.01). No differences were observed between image size ratios 1:1 (I, 0.69; II, 0.74) and 1:2 (I, 0.68; II, 0.73). Conclusions: In this study, the type of monitor did not influence approximal caries detection on digital radiographs. Image sizes with a display ratio of 1:1 and 1:2 resulted in better diagnostic validity than those with a ratio of 1:7. Dentomaxillofacial Radiology (2003) 32, 242–246. doi: 10.1259/dmfr/17654484 Keywords: digital radiography, dental; image processing; data display; dental caries Introduction Radiography is still the diagnostic standard in the detection of inaccessible approximal caries, and presently conventional dental films are frequently replaced by digital imaging systems. Monitor display of digital radiographic images differs from conventional radiographs in contrast, brightness and size. Manipulation of the grey scale allows for variation in image presentation. The increase of digital radiographic systems therefore requires recommendations on how to use these computer workplaces for diagnostic questions. Different contrast enhancement algorithms have recently been investigated regarding their influence on caries detection,1 – 3 revealing *Correspondence to: Rainer Haak, University of Cologne, Centre of Dental Medicine, Department of Operative Dentistry and Periodontology, Kerpener Str. 32, D-50937 Ko¨ln, Germany; E-mail:
[email protected] Received 4 November 2002; revised 20 May 2003; accepted 5 June 2003
task-dependent post-processing to be advantageous in compensating for visual limitations of the human observer. Relatively little information exists on the diagnostic outcome of different zooming ratios on standard monitor displays, although zooming is a routine image processing tool in digital radiography and is considered potentially advantageous.4 – 6 Møystad et al7 demonstrated that high zooming sizes gave inferior detection levels of approximal caries and that one has to be aware of an upper limit of displayed image size. Additionally, dental diagnostics based on digital images depends on the type of display source. Today, because of its predominant use in dental offices, studies in digital dental radiology are mostly confined to cathode ray tube displays, although several display parameters such as luminance and contrast ratio8 are favourable with flat panel liquid crystal monitors.9 These factors may be responsible for differences
Image size display R Haak et al
in grey-scale perception of human observers. It was demonstrated that differences in monochromatic intensities were easier to discriminate in the middle part of the grey scale when using an active matrix flat panel compared with a cathode ray tube display.10 The aim of this study was to compare the detection of primary approximal caries on radiographic images displayed on-screen at different sizes on either a cathode ray tube (CRT) monitor or a thin film transistor (TFT) monitor. It was hypothesized that a negative influence of reduced image sizes would be detected, that the use of a TFT display is advantageous and that detection of dentinal lesions results in higher accuracy compared with identification of all carious lesions. Materials and methods Specimen preparation One hundred and sixty extracted human posterior teeth, randomly collected for research purposes and stored in a 20% ethanol solution, were selected based on the following visual criteria: teeth with restorations on the approximal aspect and with carious cavitations exceeding 2 mm in diameter were excluded, leaving a disease spectrum from sound surfaces to small cavitation. Twenty pairs of upper and lower jaw quadrant blocks were mounted with approximal tooth surfaces in contact. Each block consisted of two premolars and molars with similar anatomical size and form. Canines and third molars were added with plastic teeth (AG-3; Franz Sachs & Co. GmbH, Tettnang, Germany) and the roots were embedded in plaster of Paris. No misalignments were simulated to avoid radiographic overlapping. The outline of the gingiva was simulated by a polyether mask (Vestogum; Espe, Seefeld, Germany). The models were mounted in a modified jig, while the charge-coupled device (CCD) full size sensor (Sidexisw; Sirona, Bensheim, Germany) was fixed at a focus – sensor distance of 30 cm according to the paralleling technique.3 Technical specifications of the sensor were: active area, 26 £ 34 mm2; pixel size, 19.5 £ 19.5 mm2; matrix, 880 £ 692 pixels. Soft tissue was simulated by 14 mm of dental wax placed in front of the specimen. A Heliodent DS dental X-ray unit (Sirona, Bensheim, Germany) at 60 kVp and 7 mA was used to expose the sensor for 0.08 s. Data collection Five university dentists (aged 28 – 34 years), used to routinely diagnosing digital radiographs for at least 2 years, independently examined the radiographic images of 320 approximal surfaces. The investigators indicated on a 6-point ordinal caries rating scale the presence and penetration depth of approximal carious lesions (radiolucency) (0 ¼ no lesion; 1 ¼ 1st half of enamel; 2 ¼ 2nd half of enamel; 3 ¼ 1st quarter of dentine; 4 ¼ 2nd quarter of dentine; 5 ¼ inner half of dentine). No information was given on lesion frequency. The digital images were evaluated on two display types according to a previous study10: a 1900 CRT monitor (Nokia 446 XS; Nokia Group,
243
Espoo, Finland) and a 1500 active matrix TFT display (PanaFlat LC 50 S; Panasonic GmbH, Hamburg, Germany) without protection glass or additional anti-glare coating. Screen resolution was set at 1024 £ 768 pixels; colour was set to a 16-bit depth. Visible display size of the TFT monitor was 23 £ 30.6 cm2 compared with a visible area of the CRT display of 26.2 £ 35 cm2. The images were displayed on-screen at image size ratios of 1:1, 1:2 and 1:7. At the ratio of 1:1 the radiographic images filled the screen with space left only for text and icons of the Sidexis system, and at 1:7 an extension comparable with conventional radiographs was achieved (Figure 1). The examiners used a fixed viewing distance of 70 cm with a viewing angle of 908 and were given unlimited viewing time. During the diagnostic sessions, the room light was darkened to 70 lux. Prior to each diagnostic session, contrast and brightness of the monitors were adjusted with the SMPTE test pattern.11 The sequences in which the images and display modalities were viewed were defined by random numbers, and a minimum interval of 1 week between the diagnostic sessions was kept. Histological validation and data analysis Observers’ ratings were compared with the histological extension of the approximal carious lesions. All approximal surfaces were enclosed in epoxy resin (Technovit 9100; Heraeus Kulzer, Wehrheim, Germany) and serially sectioned in a mesiodistal direction with a low speed diamond saw with a 250 mm thick diamond blade (Isomet 1000; Buehler Ltd., Lake Bluff, IL) into 300 mm sections. After reduction to approximately 80 mm by wet grinding on SiC paper, the sections were evaluated by two readers using polarized light microscopy (Axiovert 135; Carl Zeiss, Oberkochen, Germany). Disease-positive results were defined for two different validation thresholds: histological presence of caries (I) and dentine caries (II). Cases of disagreement were solved by forced consensus. One hundred and seven surfaces were free of caries, 94 exhibited caries penetration limited to enamel and 119 showed dentine lesions histologically, which resulted in frequencies of caries lesions of 67% (I) and 37% (II).
Figure 1 Sidexis desktop with simultaneous display of the same radiograph at the three image sizes under investigation (1:1, 1:2 and 1:7) Dentomaxillofacial Radiology
Image size display R Haak et al
244
Non-parametric receiver operating characteristic (ROC) analysis was calculated using the software package SPSS (SPSS Release 10.0.7; SPSS Inc., Chicago, IL) as an overall measure of accuracy of the observers’ ability to detect approximal carious lesions.12 The influences of the factors “type of display”, “image size” and “validation threshold” were examined by repeated measures analysis of variance (ANOVA) as implemented in the statistical software SAS (Release 8.00; PROC MIXED, covariance matrix: compound symmetry, post-hoc t-test; SAS Institute Inc., Cary, NC). The level for rejection of null hypotheses was set to a ¼ 0:05: As additional measures of accuracy, sensitivity and specificity were calculated for both of the validation thresholds. The caries rating scale was dichotomized in disease-positive and disease-negative scores between ratings 0 and 1 for threshold I and between 2 and 3 for threshold II.
Results ANOVA of the areas under the ROC curves (AUCs) is presented in Table 1. A significant influence on approximal caries detection was demonstrated for image size and validation threshold, whereas detection rates for both types of monitor display were not significantly different. Distributions of AUCs calculated for presence of caries as well as dentine caries are presented in Figures 2 and 3, respectively. The overlapping boxes in these illustrations confirmed similar performance for both validation thresholds on either the CRT or the TFT display. A significantly lower overall accuracy was demonstrated for the display ratio of 1:7 at both validation levels (P , 0.0001), whereas no significant differences were detected between ratios of 1:1 and 1:2 (I, P ¼ 0.96; II, P ¼ 0.53) (Table 2). AUCs based on dentine caries as disease cut-off were significantly larger compared with threshold I except for a displayed image size of 1:7 (1:1, P ¼ 0.0004; 1:2, P ¼ 0.002; 1:7, P ¼ 0.10). Sensitivity and specificity as measures of accuracy are given in Table 3. In general, based on the dichotomized data, differences between validation thresholds levelled
Figure 2 Distributions of area under the receiver operating characteristic curves (AUC) for the three display sizes (1:1, 1:2 and 1:7) grouped for monitor type for validation threshold I (presence of caries). CRT, cathode ray tube; TFT, thin film transistor
Table 1 Repeated measures ANOVA: influence of the three variables (type of display, image size and validation level) on the dependent variable AUC Effect
F-value
Significance
Type of display Image size Validation threshold Type of display £ image size Type of display £ validation threshold Image size £ validation threshold Type of display £ image size £ validation threshold
2.15 35.17 29.40 4.22
0.1497 ,0.0001 ,0.0001 0.0211
0.03
0.8538
0.55
0.5836
0.08
0.9277
AUC, area under the receiver operating characteristic curve Dentomaxillofacial Radiology
Figure 3 Distributions of area under the receiver operating characteristic curves (AUC) for the three display sizes (1:1, 1:2 and 1:7) grouped for monitor type for validation threshold II (presence of dentine caries). CRT, cathode ray tube; TFT, thin film transistor
off. If the histological standards of presence of caries (I) or dentine caries (II) were compared, a tendency towards slightly higher sensitivity combined with lower specificity could be noticed at threshold I, and vice versa (I, sensitivity ¼ 0.38, specificity ¼ 0.93; II, sensitivity ¼ 0.33, specificity ¼ 0.95). A reduction in sensitivity, which was more pronounced at threshold I, was associated with the lowest image size.
Image size display R Haak et al
Table 2
245
Means and 95% confidence intervals of area under receiver operating characteristic curve Image size
Monitor type
Validation threshold
1:1
1:2
CRT TFT Total CRT TFT Total
Caries
0.69 (0.65 –0.74) 0.68 (0.63 –0.72) 0.69 (0.64 –0.73)a 0.75 (0.71 –0.80) 0.73 (0.68 –0.77) 0.74 (0.70 –0.78)x
0.67 0.70 0.68 0.71 0.75 0.73
Dentine caries
1:7 (0.62 – 0.71) (0.66 – 0.75) (0.64 – 0.73)a (0.67 – 0.76) (0.71 – 0.80) (0.69 – 0.77)x
0.61 0.63 0.62 0.65 0.66 0.65
Total (0.57– 0.66) (0.58– 0.67) (0.58– 0.66)b (0.60– 0.69) (0.62– 0.71) (0.61– 0.70)y
0.66 (0.62 –0.70) 0.67 (0.63 –0.71) 0.66 (0.62 –0.71) 0.70 (0.66 –0.75) 0.71 (0.67 –0.76) 0.71 (0.67 –0.75)
a,b/x,y Statistical comparison (total data) of each image size: groups with same letters in a row are not significantly different at P , 0.05 CRT, cathode ray tube; TFT, thin film transistor
Table 3
Means and standard deviations of sensitivity and specificity Image size 1:1
1:2
1:7
Total
Monitor type
Validation threshold
Sensitivity
Specificity
Sensitivity
Specificity
Sensitivity
Specificity
Sensitivity
Specificity
CRT TFT Total
Caries
0.45 (0.07) 0.40 (0.13) 0.42 (0.10)
0.93 (0.04) 0.94 (0.05) 0.94 (0.05)
0.39 (0.13) 0.46 (0.12) 0.42 (0.12)
0.92 (0.02) 0.92 (0.07) 0.92 (0.05)
0.28 (0.07) 0.33 (0.07) 0.30 (0.07)
0.94 (0.02) 0.92 (0.05) 0.93 (0.03)
0.37 (0.11) 0.40 (0.07) 0.38 (0.11)
0.93 (0.03) 0.93 (0.05) 0.93 (0.04)
CRT TFT Total
Dentine caries
0.44 (0.12) 0.33 (0.14) 0.38 (0.13)
0.94 (0.04) 0.96 (0.03) 0.95 (0.03)
0.33 (0.18) 0.32 (0.13) 0.32 (0.13)
0.95 (0.03) 0.96 (0.03) 0.96 (0.03)
0.27 (0.10) 0.31 (0.09) 0.29 (0.09)
0.94 (0.02) 0.92 (0.03) 0.93 (0.03)
0.34 (0.15) 0.32 (0.11) 0.33 (0.13)
0.94 (0.03) 0.95 (0.03) 0.95 (0.03)
CRT, cathode ray tube; TFT, thin film transistor
Discussion Display of radiographic images using electronic devices including CRT or flat-panel technology is rapidly becoming more common within dental radiology. However, digital image presentation is different to conventional systems, first, in brightness, contrast and size, and second, for different monitor technology. Studies comparing examples of different monitor types in caries detection were, however, not able to reveal relevant differences.13 – 15 This is consistent with the observations in this study, but contradicted our initial alternative hypothesis. We expected that the advantages of the TFT display in greyscale perception in the middle of the grey-scale range10 could optimize approximal caries detection. It could be stated that the flat panel display was equally suited for approximal caries detection compared with the CRT monitor. Whilst magnification of radiographic films is limited to the use of magnifying lenses, image sizes on monitor displays can be modified across a wide range. How large an image appears on-screen depends on the combination of different factors such as pixel dimension of the image, monitor size and resolution setting of the display. Increasing the image size above ratios of 1:1 results in pixelation with coarse-looking image details and it was reported that the display of bitewing radiographs extending £ 18 and £ 30 that of conventional film gave inferior detection rates of approximal carious lesions.7 The comparability between the reported magnification levels and that in our study could only be estimated. Møystad et al7 digitized conventional radiographs to the dimension of 752 £ 582 pixels and displayed the images
on a 1700 monitor at a resolution of 800 £ 600 pixels, which resulted in nearly full-screen display at an image size of 1:1. However, the magnification levels ( £ 3, £ 6, £ 12, £ 18, £ 30) were related to the standard film size (4 £ 3 cm2). Provided that the multiplier was related to the distances and not to the area of the conventional radiograph, a nearly £ 9 magnification would be equivalent to a full-screen image display. Excessive increase of the image size with coarse-looking image pixels could serve as an explanation for the reduced caries detection rates. The AUCs of the £ 3 (AUC ¼ 0.70), £ 6 (AUC ¼ 0.70) and £ 12 (AUC ¼ 0.73) magnification are in accordance with our results for the display ratios of 1:1 and 1:2. It might be misleading to discuss the image sizes on-screen generated by the digital zoom function using the term “magnification”. Above a display ratio of 1:1, no further information is added but only the pixel size increases, and below 1:1 the visualized information is dramatically reduced. On a typical monitor with a maximum resolution of 96 dpi, an area corresponding to a film radiograph allows for approximately 17 000 pixels, whilst a digital system (e.g. Sidexis) supplies 608 960 (880 £ 692) pixels of information. The effect of not considering the available radiographic information was seen for the image size of 1:7, which demonstrated a lower detection rate of approximal carious lesions. Svanæs et al16 also reported a lower diagnostic validity for this scenario if Digora images are displayed on an area of 3 £ 4 cm2 with a resolution of 1024 £ 768 compared with an image size of 12 £ 16 cm2, which corresponds to our display ratio of 1:1. The negative effect of zooming out in detecting smaller Dentomaxillofacial Radiology
Image size display R Haak et al
246
details was also seen for the radiographic assessment of endodontic files in length determination.17,18 The practical consequences of using reduced digital image display could be seen by calculating the predictive values. If one assumes a low caries prevalence of 10%, the reduction in sensitivity and specificity between image size 1:1 and 1:7 for validation threshold I results in a positive predictive value (PPV) of 43% compared with 32%, and negative predictive values of 94% and 92%, respectively (II: PPV ¼ 46% vs 32%; NPV ¼ 93% vs 92%). In conclusion, the present study indicates that the chosen monitor types did not influence approximal caries
detection on digital radiographs and that the TFT display performed at least as well as the CRT display. Image sizes with a display ratio of 1:1 and 1:2 resulted in higher diagnostic validity and so it is recommended to use an image size displaying the complete pixel information of the digital system used.
Acknowledgment We wish to thank Sirona Dental Systems GmbH for providing the Sidexis system.
References 1. Verdonschot EH, Kuijpers JMC, Polder BJ, de Leng-Worm MH, Bronkhorst EM. Effects of digital grey-scale modification on the diagnosis of small approximal carious lesions. J Dent 1992; 20: 44 – 49. 2. Eickholz P, Kolb I, Lenhard M, Hassfeld S, Staehle H. Digital radiography of interproximal caries: effect of different filters. Caries Res 1999; 33: 234 –241. 3. Haak R, Wicht MJ, Noack MJ. Conventional, digital and contrastenhanced bitewing radiographs in the decision to restore approximal carious lesions. Caries Res 2001; 35: 193 –199. 4. Paakkala T, Aalto J, Ka¨ha¨ra¨ V, Seppa¨nen S. Diagnostic performance of a teleradiology system in primary health care. Comput Methods Programs Biomed 1991; 36: 157 – 160. 5. Furkart AJ, Dove SB, McDavid WD, Nummikoski P, Matteson SR. Direct digital radiography for the detection of periodontal bone lesions. Oral Surg Oral Med Oral Pathol 1992; 74: 652 – 660. 6. Wenzel A, Pitts N, Verdonschot EH, Kalsbeek H. Developments in radiographic caries diagnosis. J Dent 1993; 21: 131 –140. 7. Møystad A, Svanæs DB, Larheim TA, Gro¨ndahl HG. Effect of image magnification of digitized bitewing radiographs on approximal caries detection: an in vitro study. Dentomaxillofac Radiol 1995; 24: 255 –259. 8. Dwyer SJ 3rd, Stewart BK, Sayre JW, Aberle DR, Boechat MI, Honeyman JC, et al. Performance characteristics and image fidelity of gray-scale monitors. Radiographics 1992; 12: 765 – 772. 9. Flynn M, Kanicki J, Badano A, Eyler W. High-fidelity electronic display of digital radiographs. Radiographics 1999; 19: 1653– 1669. 10. Haak R, Wicht MJ, Hellmich M, Nowak G, Noack MJ. Influence of room lighting on grey-scale perception with a CRT- and a TFT monitor display. Dentomaxillofac Radiol 2002; 31: 193 –197.
Dentomaxillofacial Radiology
11. Society of Motion Pictures and Television Engineers (SMPTE). Specifications for medical diagnostic imaging test pattern for television monitors and hard-copy recording cameras. Recommended Practice RP 133-1986. SMPTE Journal 1986; 95: 693 – 695. 12. Swets JA, Pickett RM. Evaluation of diagnostic systems: methods from signal detection theory. New York, NY: Academic Press, 1982. 13. Abreu M Jr, Tyndall DA, Ludlow JB. Detection of caries with conventional digital imaging and tuned aperture computed tomography using CRT monitor and laptop displays. Oral Surg Oral Med Oral Pathol Oral Radiol Endod 1999; 88: 234 –238. 14. Cederberg RA, Frederiksen N, Benson BW, Shulman JD. Influence of the digital image display monitor on observer performance. Dentomaxillofac Radiol 1999; 28: 203 –207. 15. Ludlow JB, Abreu M Jr. Performance of film, desktop monitor and laptop displays in caries detection. Dentomaxillofac Radiol 1999; 28: 26 –30. 16. Svanæs DB, Møystad A, Risnes S, Larheim TA, Gro¨ndahl HG. Intraoral storage phosphor radiography for approximal caries detection and effect of image magnification. Oral Surg Oral Med Oral Pathol Oral Radiol Endod 1996; 82: 94 – 100. 17. Ellingsen MA, Hollender LG, Harrington GW. Radiovisiography versus conventional radiography for detection of small instruments in endodontic length determination. II. In vivo evaluation. J Endodont 1995; 21: 516 –520. 18. Versteeg CH, Sanderink GCH, Lobach SR, van der Stelt PF. Reduction in size of digital images: does it lead to less detectability or loss of diagnostic information? Dentomaxillofac Radiol 1998; 27: 93 – 96.
