Dentomaxillofacial Radiology (2002) 31, 113 ± 116 ã 2002 Nature Publishing Group. All rights reserved 0250 ± 832X/02 $25.00 www.nature.com/dmfr
RESEARCH
Detection of approximal caries in digital radiographs before and after correction for attenuation and visual response. An in vitro study G Li*,1, K Yoshiura1,2, U Welander1, X-Q Shi1 and WD McDavid3 1
Department of Oral Radiology, Karolinska Institutet, Stockholm, Sweden; 2Department of Oral and Maxillofacial Radiology, Graduate School of Dental Science, Kyushu University, Fukuoka, Japan; 3Department of Dental Diagnostic Science, University of Texas Health Science Center at San Antonio, San Antonio, Texas, USA
Objective: To evaluate if digital compensation for exponential attenuation and the characteristics of the human visual system improves the diagnosis of approximal caries from digital radiographs. Material and methods: Forty premolar teeth were mounted in plaster blocks. Radiographs of the teeth were exposed with the Dixi digital intraoral system employing a Prostyle Intra dental X-ray unit (Planmeca Oy, Helsinki, Finland). Thirteen radiographs were then processed to compensate for the exponential attenuation and for the characteristic of the human visual system using equations presented in the paper. Ten observers were asked to diagnose approximal caries in all radiographs. ROC analyses were performed. The teeth were subsequently sectioned for histological validation of the lesions. The areas under ROC curves of original and processed radiographs were compared and analysed using Wilcoxon's signedranks test. Results: There were signi®cant diagnostic dierences between the two types of radiographs (all lesions P=0.005, enamel P=0.028, and dentine P=0.050). Conclusion: Digital radiographs processed to compensate for exponential attenuation and the characteristics of the human visual system signi®cantly improves the diagnosis of approximal caries in vitro. Dentomaxillofacial Radiology (2002) 31, 113 ± 116. DOI: 10.1038/sj/dmfr/4600675 Keywords: radiography, dental; radiography, dental, digital; dental caries, approximal; ROC curve; image processing, computer-assisted Introduction Digital radiographs have become widely accepted and a number of publications report that the diagnostic accuracy of digital radiographs is at least as good as that of conventional ®lm radiographs.1,2 To improve the diagnostic accuracy of digital radiographs in caries diagnosis, dierent image processing techniques have been investigated, such as contrast and brightness adjustment, histogram equalization, noise reduction, and pseudocoloring.3 ± 10 Applied in an optimum manner, these methods may improve the diagnosis of carious lesions. Finding convenient and ecient methods to improve caries diagnosis is still a common objective among
*Correspondence to: G Li, Oral Radiology, Karolinska Institutet, Box 4064, 141 04 Huddinge, Sweden; E-mail:
[email protected]. Received 26 June 2001; revised 11 October 2001; accepted 6 December 2001
researchers. Displaying digital radiographs in such a way that equal steps in object thickness will give equal steps in gray levels may be one way of improving caries diagnosis. In a recent study, a method of achieving this goal was presented.11 An additional linearization was also performed so that equal steps in gray levels will be perceived as equal steps in output brightness from a computer monitor.12 It was shown that this combined approach improves the detection from radiographs of holes of dierent depths in a step wedge test object.11 This method of linear steps in intensity of radiographic information improves perception and could therefore be expected to improve caries diagnosis. The aim of the present work was to determine if linearization of digital radiographs to compensate for exponential attenuation and the characteristics of the
Digital caries detection G Li et al
114
human visual system improves diagnosis of approximal caries. The null hypothesis was stated as `there is no dierence between the modalities' and rejected when P50.05. Material and methods The investigation was performed as an in vitro study. Teeth The material consisted of 40 premolars extracted from young adolescents in the course of orthodontic treatment. All teeth that were available were included in the material without employing speci®c selection criteria. The teeth were mounted in plaster blocks. The most prominent parts of the approximal surfaces were at the same vertical level to simulate their anatomical positions. The blocks of teeth were then numbered and stored separately in 10% neutral buered formalin solution. Test radiographs The Dixi digital intraoral system (Planmeca Oy, Helsinki, Finland) was used to expose radiographs of the blocks of teeth employing a Prostyle Intra dental unit (Planmeca Oy, Helsinki, Finland). The exposure time was 0.20 s, the nominal tube potential was 63 kVp, and the tube current was 8 mA. The focusdetector distance was 21 cm. To assure reproducible radiographs, the blocks of teeth were placed in a specially designed holder mounted on the cone of the X-ray tube, allowing standard projection geometry. A 15 mm thick soft tissue equivalent plastic compound was placed in front of the tooth blocks. Thirteen digital radiographs were exposed. A previous study has shown that optimum caries diagnosis is achieved when the exposure gives a gray level in enamel and dentine that corresponds to the middle light intensity output from a computer monitor, i.e. an average value of about 127 when 8 bit data are used for the display.13 We therefore exposed the test radiographs in order to achieve mean gray levels in enamel and dentin of about 127, i.e. in the middle range of the brightness scale of a monitor. This should be an ideal exposure in a clinical situation.13 The 13 original radiographs were then processed to obtain linearization to correct for the exponential attenuation and the response of the eye employing methods described in previous papers.11,12 Thus, the material totaled 26 radiographs, 13 originals and 13 recalculated radiographs. These radiographs were combined into one series and the order randomized. Viewing Ten observers, all dentists and experienced in caries diagnosis, evaluated all the test radiographs in one session. The radiographs were displayed on the Nokia
Dentomaxillofacial Radiology
447R monitor (Nokia, Salo, Finland). Prior to the viewing, brightness and contrast were optimized and pre-set by the investigators using the Nokia Screen Test Program 1.0. Viewing took place in an arti®cially lit room. No information about the number of carious lesions was given to the observers. Observers were instructed to record their level of con®dence about the presence or absence of carious lesions on approximal surfaces according to the following scale: 1. de®nitely not caries; 2. probably not caries; 3. questionable; 4. probably caries; 5. de®nitely caries.
Histological analysis The teeth were prepared for histological examination. In order to isolate each approximal site on each specimen, the teeth were embedded in methylmethacrylate and hemi-sectioned with a diamond saw perpendicular to the occlusal and buccal surfaces. Each half of the tooth crown was then re-embedded in methylmethacrylate. At the site of maximum depth of the lesion, identi®ed by direct smooth surface measurement by DIAGNOdent,14 300 mm thick slices were sawn perpendicular to the enamel surface. The tooth slices were examined under a microscope at 166 magni®cation. Two observers, both experienced in histological examination, evaluated the extension of the lesions together. The depths of the lesions were de®ned by a whitish decalci®ed zone and/or a brown zone toward the pulp. A three-point scale was used for evaluating the depth of carious lesions: sound, enamel caries, and dentine caries. ROC analysis Observer performance for original and processed radiographs was analysed using the histological examination as the gold standard by Receiver Operating Characteristic (ROC) curves10,15,16 with program `ROCKIT 0.9B, Beta version' (University of Chicago, Chicago, IL, USA).The maximum likelihood parameters were found and the area, Az, under each ROC curve were calculated. Statistical analysis Az values from each observer were used for statistical analyses employing Wilcoxon's signed-ranks test. Results The status of the 80 approximal surfaces in the teeth of the blocks are presented in Table 1. Six specimens were lost during the preparation of tooth slices.
Digital caries detection G Li et al
115
Table 1 Histological validation of the tooth material
a Sound Enamel caries Dentine caries Total
No of tooth surfaces
Per cent
12 43 19 74
16 58 26 100
Table 2 Mean areas, Az, under the ROC curves for all observers. Lesion type Enamel Dentine Enamel + Dentine
b
Az, original radiographs
Az, processed radiographs
0.533 0.842 0.575
0.610 0.903 0.660
Figure 1 illustrates ROC curves for enamel, dentin and all carious lesions for original and recalculated image. The curves represent combined data from the 10 observers. The ROC curves for the processed radiographs are higher than those for original and the areas under the ROC curves, Az, are larger. Az values are listed in Table 2. Wilcoxon's signed-ranks test gave P-values of 0.005 for all carious lesions, 0.028 for enamel lesions and 0.050 for dentinal lesions. This result indicates that the dierences between original and processed radiographs were signi®cant for enamel lesions and very close to signi®cant for dentinal lesions. Discussion
c
Figure 1 ROC curves for original radiographs and processed radiographs. (a) enamel carious lesions, (b) dentinal carious lesions, (c) all carious lesions
With the introduction of a new digital image processing method the ®rst question that needs to be answered is whether or not the new method improves diagnosis. ROC analysis is an established method that can be used to answer this question. The present work shows that processing digital radiographs to correct for the exponential attenuation and to achieve visual linearization may have advantages in caries diagnosis. The diagnostic accuracy as de®ned by the Az values was low for enamel lesions: 0.533 for originals and 0.610 for processed radiographs. In fact, these values are quite typical for enamel lesions.17 The corresponding values for dentin were high: 0.842 and 0.903, respectively. Values for all carious lesions were low: 0.575 and 0.660, respectively. The explanation for these low values is most certainly that in the case sample employed in this study there was an over representation of carious lesions in the enamel. However, this should not aect the overall conclusion that indicates that the type of digital processing advocated here seems to have bene®ts. In our opinion, the present results indicate that digital radiographs should be corrected for X-ray attenuation and the response on of the human visual system to light intensities when viewing radiographs on Dentomaxillofacial Radiology
Digital caries detection G Li et al
116
a cathode ray tube computer screen. This setting should be the point of departure for further digital image processing. However, future studies should investigate this problem further.
In conclusion, correction of digital radiographs for the exponential attenuation combined with visual linearization seems to improve caries diagnosis.
References 1. White SC, Yoon DC. Comparative performance of digital and conventional images for detection proximal surface caries. Dentomaxillofac Radiol 1997; 26: 32 ± 38. 2. Syriopoulos K, Sanderink GCH, Velders XL, Stelt PF. Radiographic detection of approximal caries: a comparison of dental ®lms and digital imaging systems. Dentomaxillofac Radiol 2000; 29: 312 ± 318. 3. Pitts NB. Detection and measurement of approximal radiolucencies by computer-aided image analysis. Oral Surg Oral Med Oral Pathol 1984; 58: 358 ± 366. 4. Wenzel A, Hintze H, Mikkelsen L, Mouyen F. Radiographic detection of occlusal caries in noncavitated teeth. A comparison of conventional ®lm radiographs, digitized ®lm radiographs, and RadioVisioGraphy. Oral Surg Oral Med Oral Pathol 1991; 72: 621 ± 626. 5. Wenzel A. Computer-aided image manipulation of intraoral radiographs to enhance diagnosis in dental practice: a review. Int Dent J 1993; 43: 99 ± 108. 6. Wenzel A, Hintze H. Perception of image quality in direct digital radiography after application of various image treatment ®lters for detectability of dental disease. Dentomaxillofac Radiol 1993; 22: 131 ± 134. 7. Shrout MK, Russell CM, Potter BJ, Powell BJ, Hideboldt CF. Digital enhancement of radiographs: Can it improve caries diagnosis? J Am Dent Assoc 1996; 127: 469 ± 473. 8. Moystad A, Svanaes DB, Risnes S, Larheim TA, Grondahl HG. Detection of approximal caries with a storage phosphor system. A comparison of enhanced digital images with dental X-ray ®lm. Dentomaxillofac Radiol 1996; 25: 202 ± 206.
Dentomaxillofacial Radiology
9. GoÈtfredsen E, Wenzel A, Grondahl HG. Observers' use of image enhancement in assessing caries in radiographs taken by four intra-oral digital systems. Dentomaxillofac Radiol 1996; 25: 34 ± 38. 10. Svanaes DB, Moystad A, Larheim TA. Approximal caries depth assessment with storage phosphor versus ®lm radiography. Evaluation of the caries-speci®c Oslo enhancement procedure. Caries Res 2000; 34: 448 ± 453. 11. Welander U, Yoshiura K, Li G, SaÈllstroÈm P, McDavid WD. Correction for x-ray attenuation in digital radiographs for visual linearization. Dentomaxillofac Radiol 2002; 31: 117 ± 125. 12. Li G, Yoshiura K, Welander U, SaÈllstroÈm P, McDavid WD. Visual linearization of the display of digital radiographs. Dentomaxillofac Radiol 2002; 31: 131 ± 136. 13. Svenson B, Shi X-Q, Stamatakis H, Welander U, Soderfeldt B, Tronje G et al. Exposure parameters in Sens-A-Ray direct digital radiography and their eects on diagnostic accuracy of carious lesions. Oral Radiol 1997; 2: 83 ± 91. 14. Shi XQ, Tranaeus S, Angmar-MaÊnsson B. Validation of DIAGNOdent for quanti®cation of smooth surface caries: an in vitro study. Acta Odontol Scand 2001; 59: 74 ± 78. 15. Swets JA. ROC analysis applied to the evaluation of medical imaging techniques. Invest Radiol 1979; 14: 109 ± 121. 16. Shi XQ, Han P, Welander U, Angmar-MaÊnsson B. Tunedaperture computed tomography for detection of occlusal caries. Dentomaxillofac Radiol 2001; 30: 45 ± 49. 17. Wenzel A, Hintze H. The choice of gold standard for evaluating tests for caries diagnosis. Dentomaxillofac Radiol 1999; 28: 132 ± 136.
