Original Paper Caries Res 2007;41:121–128 DOI: 10.1159/000098045
Received: March 24, 2005 Accepted after revision: August 9, 2006
Accuracy and Reproducibility of Conventional Radiographic Assessment and Subtraction Radiography in Detecting Demineralization in Occlusal Surfaces D.N.J. Ricketts a K.R. Ekstrand a, c S. Martignon c, d R. Ellwood b M. Alatsaris a Z. Nugent a a c
University of Dundee Dental School, Dundee, and b Dental Health Unit, Manchester, UK; University of Copenhagen, Copenhagen, Denmark; d University of El Bosque, Bogota, Colombia
Key Words Digital radiography Occlusal caries Subtraction
Abstract Subjective interpretation of paired digital radiographic images viewed side by side to assess occlusal lesion progression, arrest or remineralization is difficult. The aim of this study was to compare the accuracy and reproducibility of a digital subtraction radiography technique and visual assessment of paired digital images in detecting changes in mineral content within occlusal cavities. Forty molar teeth with occlusal cavities were placed in arches and baseline digital radiographs taken. Nineteen teeth were randomly selected and had acid placed in the cavities and digital images taken after 3, 6, 12, 18 and 24 h of acid exposure. Paired baseline images and those taken at the various time intervals were examined side by side and assessed for demineralization by five examiners. Subtraction images prepared from the paired images were assessed in the same way. One fifth of the images were re-examined to determine intra-examiner reproducibility. After 12 h or longer the diagnostic accuracy (mean area under the ROC curve = 0.92–0.98 for subtraction radiography), intra-examiner and inter-examiner reproducibility for detection of demineralization from the subtraction im-
© 2007 S. Karger AG, Basel 0008–6568/07/0412–0121$23.50/0 Fax +41 61 306 12 34 E-Mail
[email protected] www.karger.com
Accessible online at: www.karger.com/cre
ages was significantly better than viewing the paired images side by side (p ! 0.01). The subtraction radiography system used was found to be more accurate and reproducible than visual assessment of paired digital images. As such the technique shows promise for monitoring occlusal lesion progression in clinical studies. Copyright © 2007 S. Karger AG, Basel
The presentation and management of dental caries have changed dramatically over the last three to four decades as caries prevalence has declined. Fluoride has had a profound effect on the incidence of smooth surface caries, so that occlusal caries now accounts for the majority of new lesions in children, adolescents and young adults [Dummer et al., 1988; Ripa et al., 1988]. Topical fluoride may make occlusal caries detection more difficult by making superficial enamel harder, thus allowing lesions to spread before any obvious cavitation occurs. Such lesions are often missed from a visual examination, but are extensive enough to be detected radiographically, and terms such as ‘hidden caries’ or ‘fluoride caries’ have been coined [Ricketts et al., 1997]. However, evidence now suggests that meticulous examination of clean teeth, wet and dry, can eliminate this apparent phenomenon by relating
Dr. David N.J. Ricketts Dundee Dental School, Park Place Dundee, DD1 4HR (UK) Tel./Fax +44 1382 635 984 E-Mail
[email protected]
visual appearance at the entrance to pits and fissures to the histopathology of the disease [Ekstrand et al., 1997, 1998]. Caries management has also changed with greater emphasis on preventive treatment of early lesions. The aim of such treatment is to disturb the plaque or biofilm with improved oral hygiene, combined with fluoride tooth paste to promote remineralization [Carvalho et al., 1992]. More extensive occlusal lesions which have been missed from a visual examination, but are deep enough to be detected radiographically, have also been treated with minimally invasive techniques such as fissure sealing [MertzFairhurst et al., 1998]. The aim again is microbial disruption, this time by depriving organisms within the carious dentine of sugar substrate from the oral cavity. If such approaches to caries management are adopted, recall and monitoring are essential. New detection devices have been introduced to detect and monitor early lesions with objective measurements, based upon electrical conductance measurements and laser-induced fluorescence, for example [Longbottom and Huysmans 2004; Ricketts et al., 1995; Stookey et al., 1999]. In the situation where a larger occlusal lesion has been fissure sealed, such diagnostic tools are not appropriate and the only way to monitor the lesion is by bitewing radiographs. However, accurate assessment of occlusal lesion extent from a conventional bitewing radiograph is unlikely [Ricketts et al., 1994] and once a radiograph is developed, little can be done to improve its diagnostic quality. As an alternative to conventional radiography, there are two main types of digital imaging systems: real time (corded) and photo-stimulable phosphor storage plates (cordless). The corded system’s sensor is relatively small in area and bulky, which makes positioning difficult and retake images likely. The phosphor plates on the other hand are produced in the same sizes as conventional intra-oral radiographs and are prone to less positional errors than the corded sensors [Bahrami et al., 2003]. Using direct digital radiography, if two digital images are taken with the same X-ray projection geometry, it is possible with specially created computer programs to superimpose one image upon another and subtract the grey scale values, resulting in an image representing the difference between them (subtraction image). If two images are identical, this would produce an image with no detail. However, if a carious lesion has progressed, the difference is seen as either an increase or decrease in pixel values on a more uniform background. By increasing the contrast in this image the caries signal is magnified so that changes can be detected more easily [Eberhard et al., 2000; 122
Caries Res 2007;41:121–128
Wenzel et al., 1993]. In practice, the method has limitations, not least of which is the requirement to maintain projection geometries as uniform as possible when capturing both radiographs. Whilst there are a number of publications investigating subtraction radiography, most have been carried out on bone; relatively few have been carried out on caries and those that have, have mainly investigated approximal surfaces [Eberhard et al., 2000; Nummikoski et al., 1992; Schmidlin et al., 2002]. This may be due to the fact that changes in optical density on a digital image are more readily detected in less opaque tissues such as bone when the signal-to-noise ratio is lower than found in enamel. In some subtraction studies, artificial ‘lesions’ are created by drilling a hole in the proximal surface of a tooth [Nummikoski et al., 1992; Schmidlin et al., 2002]. This produces a ‘lesion’ that has well-demarcated boundaries and is more readily detectable. Limited data are available on the use of subtraction radiography of occlusal caries and the only known study that has, has investigated the remineralizing effect of stannous fluoride treatment [Wenzel and Halse, 1992]. The aim of this study was therefore to compare the accuracy and reproducibility of a digital subtraction radiography technique and visual assessment of paired digital images in detecting changes in mineral content (demineralization) within occlusal cavities.
Materials and Methods Teeth Selected Forty extracted molar teeth were selected which had unrestored occlusal surfaces with a range of visual appearances from sound to frank cavitation. The teeth were cleaned and stored throughout the study in saline containing thymol crystals to inhibit any further bacterial activity. Two molar teeth were selected at a time and mounted along with one premolar tooth in plaster to simulate the anatomical relationship of upper and lower posterior sextants. Only the roots of the teeth were invested in plaster up to 3–4 mm short of the cement-enamel junction. Pink dental modelling wax was then placed to the cement-enamel junction and contoured to simulate the gingival soft tissues. In total 20 posterior sextants were created, 5 with upper left teeth, 5 with lower left teeth, 5 with lower right teeth and 5 with upper right teeth. The plaster blocks were trimmed to allow close approximation of the phosphor image plates to the lingual surface of the teeth. For each molar tooth, the fissures were opened with a diamond bur to expose the sound or carious dentine beneath. In the sound teeth the cavities cut were 2–3 mm wide in a bucco-lingual direction, and in the teeth with dentine caries, only the enamel over the carious dentine was removed.
Ricketts /Ekstrand /Martignon /Ellwood / Alatsaris /Nugent
Line drawn on laminated sheet to allow alignment of x-ray tube
Phosphor plate
a Teeth in plaster blocks
Bench top
Perspex soft tissue equivalent
X-ray tube
Phosphor plate
b
Fig. 1. Aerial (a) and horizontal (b) view of
X-ray projection geometry setup.
Radiographic Technique Baseline direct digital radiographic images were taken using an alignment system which would allow reproducible X-ray projection geometry at subsequent radiographic examination (fig. 1). This consisted of a laminated board which was placed on a horizontal level work surface, with its leading edge aligned with the edge of the work surface. On the laminated board, two adjacent and parallel rectangles were marked for sighting of the plaster block with the teeth and a 10-mm thick Perspex block, which was used as a soft tissue equivalent. The front of the soft tissue equivalent block was placed at the leading edge of the laminated board. The phosphor digital imaging plates (DenOptixTM photo-stimulable phosphor digital imaging system), contained within their light protective sleeves, were attached to the plaster blocks in relation to the lingual aspect of the teeth. The blocks were placed such that the buccal surfaces of the teeth were facing the X-ray beam (fig. 1). A line was drawn at right angles to the plaster block holding the teeth and soft tissue equivalent, allowing alignment of the X-ray tube. The vertical angulation of the X-ray beam was set at 0 degree to the horizontal plane, and the end of the X-ray tube was placed directly against the soft tissue equivalent. Using this X-ray projection geometry, radiographs were taken using a Gendex 765DC X-ray source (65 kV, 7 mA, exposure time 0.16 s, sourceto-detector distance 250 mm). The latent images on the phosphor plates were then read using a DenOptix scanner and VixWin software. The images obtained were stored as uncompressed TIF images (367 ! 485 dpi, 256 grey scale) and resized to 454 ! 600 dpi (original scanned size). Digital subtraction [Ellwood et al., 1998] was undertaken using Compare software (Dental Health Unit, University of Manchester, UK) which runs as a plug-in to the Image Tool version 1.23 software (University of San Antonio, Texas).
Demineralization of Teeth Nineteen of the forty molar teeth were randomly selected by one of the authors (K.E.) to receive a demineralization solution (pH = 1; Surgipath Decalcifier II), which was blind to the examiners. A pipette was used to drop the demineralization solution into the test cavities which was continually replenished throughout the study. After 3, 6, 12, 18 and 24 h the acid was rinsed out of the cavities with water and digital radiographs were taken using the same procedure as for the baseline radiographs. Once the radiograph had been taken, more demineralization solution was placed in the test cavities.
Detection of Occlusal Caries Using Subtraction Radiography
Caries Res 2007;41:121–128
Image Manipulation and Observations The digital images were then processed by another of the authors (S.M.) using the Image Tool Compare module. In total, images were obtained for 20 posterior sextants at 6 time intervals, including baseline, giving 120 images each with 2 molar teeth. From this point the image of each molar tooth was considered separately, giving a total of 240 subtraction images or 240 paired digital images of individual teeth. Subtraction images were created by subtracting baseline images from baseline, 3-hour demineralization images from baseline, 6-hour images from baseline and like wise for 12-, 18- and 24-hour demineralization images. In the subtraction programme the images were first automatically aligned using a patch minimization process to identify the relationship between two images with the least difference between them. In order to facilitate the subtraction process, differences in brightness and contrast between two paired images were corrected using a regression of pairs of pixels at the same image address in the two images. This transformation was then applied to the second image using the model coefficients. Finally the images were subtracted and 128 grey levels added to the image to move the values into the mid range of the grey scale values. The
123
Fig. 2. At the top of each column is the
baseline digital image, the one below is the image after possible demineralization with acid for the two teeth arrowed. This is how the images would have been viewed side by side. The bottom image is the corresponding subtraction image. The subtraction image on the left shows no demineralization, whilst the subtraction image on the right shows clear mineral loss.
contrast in the images was then increased using the Image Tool slider to magnify the caries signal. The subtraction images (total n = 240) were coded and randomly saved to 5 equally sized Microsoft Powerpoint presentations. A random numbers table was used for randomization [Daniel, 1974]. Powerpoint presentations of paired digital images (n = 240) were also created in a similar manner, so that the corresponding paired images could be viewed side by side. The Powerpoint presentations were viewed by five examiners, blind to the randomization of which teeth had received the demineralization solution and which time periods were being compared. All five examiners were qualified dentists (12 years) interested in caries diagnosis but none were specialist radiologists. The examiners were from different countries and Powerpoint presentations were used for convenient distribution. The monitors used to view the digital images and the ambient lighting conditions were not standardized. However, the viewing conditions were those used routinely by each examiner. Each presentation was viewed on separate occasions to reduce examiner fatigue. From each subtraction image or pair of digital images one of the following decisions was made: 1 Definitely no demineralization 2 Likely no demineralization 3 Do not know 4 Likely demineralization 5 Definite demineralization One fifth of all subtraction and paired digital images were randomly selected and re-examined by each examiner using the same procedure.
124
Caries Res 2007;41:121–128
Statistical Analysis Using the placement of the demineralizing solution as the gold standard for validating demineralization within occlusal cavities, sensitivity and specificity were calculated at each diagnostic threshold for viewing subtraction images and paired digital images, viewed side by side, at each time period with the solution and for each examiner. From these data receiver operator characteristic (ROC) curves were created and the area under the curves calculated for each examiner using both examination techniques and after each time interval (ROC program SPSS). The areas under the ROC curves were compared using analysis of variance with a split plot design in which the different examiners constituted plots. Intra-examiner reproducibility was determined using statistics. The following classification was used: ^ 0.20 poor agreement, 0.21 ^ ^ 0.40 fair agreement, 0.41^ ^ 0.60 moderate agreement, 0.61 ^ ^ 0.80 substantial agreement and 1 0.80 good agreement [Landis and Koch, 1977]. values for inter-examiner reproducibility were calculated for pairs of examiners. In total there were 10 inter-examiner comparisons for each examination technique after each time period with acid.
Results
Figure 2 shows two sets of digital images as viewed, side by side during the Powerpoint presentations and the corresponding subtraction images. From the subtraction image on the left it is clear that no demineralization has taken place, as the images completely subtract and no imRicketts /Ekstrand /Martignon /Ellwood / Alatsaris /Nugent
Table 1. The areas under the ROC curves generated for the assessment of demineralization within occlusal
cavities after various time periods with a demineralising solution Time period, h
Method: Digital vs. digital Examiner: 1
0 3 6 12 18 24
0.49 0.57 0.67 0.76 0.85 0.82
Subtraction
2
3
4
5
1
2
3
4
5
0.51 0.49 0.49 0.39 0.57 0.74
0.51 0.56 0.56 0.61 0.67 0.83
0.45 0.57 0.57 0.60 0.62 0.64
0.55 0.57 0.60 0.67 0.56 0.77
0.61 0.64 0.76 0.95 0.99 1.00
0.59 0.53 0.74 0.97 1.00 0.92
0.53 0.70 0.74 0.92 0.94 0.94
0.61 0.72 0.74 0.94 0.95 0.95
0.55 0.62 0.59 0.98 1.00 0.97
Areas were determined by five examiners viewing digital radiographs side by side and by subtraction radiography images.
Table 2. The main effects and
interactions between the examiners, method of viewing the radiographs and the time with potential acid (ANOVA)
Source of variation
SS
d.f.
MS
F
p
Examiner (A) Method (B) AB (whole plot error) Time (C) AC BC ABC (subplot error)
0.0583 0.5536 0.0346 0.8388 0.0628 0.1680 0.0787
4 1 4 5 20 5 20
0.01457 0.55357 0.00866 0.16775 0.00314 0.03361 0.00394
3.70 63.92
0.02 0.0013
53.45
