tomography (N=8), and linear tomography (N=8). Eight trained observers were asked to identify the presence or absence of caries on each surface using a five ...
Dentomaxillofacial Radiology (1998) 27, 193 ± 198 1998 Stockton Press All rights reserved 0250 ± 832X/98 $12.00 http://www.stockton-press.co.uk/dmfr
Extraoral radiographic imaging of primary caries TL Clifton1, DA Tyndall2 and JB Ludlow2 1
Pediatric Dentistry and Orthodontics Graduate Programs; 2Diagnostic Sciences Department, University of North Carolina School of Dentistry, Chapel Hill, North Carolina, USA
Objective: To evaluate three extra-oral radiographic imaging modalities as alternatives to conventional intra-oral ®lm for the detection of primary caries. Methods: Sixty-four extracted primary molar teeth with eighty-®ve carious lesions were radiographed using D-speed ®lm (N=8), panoramic imaging (N=8), multidirectional tomography (N=8), and linear tomography (N=8). Eight trained observers were asked to identify the presence or absence of caries on each surface using a ®ve point scale. Ground sections were viewed microscopically to determine truth. ROC curve areas (AZ) were generated from observer responses and assessed with ANOVA. Results: Averages of AZ for the detection of combined results for proximal and occlusal lesions were 0.70 for D-speed ®lm, 0.58 for linear tomography, 0.64 for both multidirectional tomography, and panoramic ®lm. Tukey's pairwise comparisons of AZ revealed that D-speed ®lm was signi®cantly better than linear tomography (P=0.0039). When data were divided into proximal and occlusal surfaces the variability due to modality remained signi®cant (P=0.0003 and P=0.0024 respectively). Tukey's comparisons for proximal surfaces revealed that D-speed ®lm was signi®cantly better than linear tomography (P=0.0007), multidirectional tomography (P=0.0010) and panoramic radiography (P=0.0100). For detection of occlusal lesions, multidirectional tomography was signi®cantly better than linear tomography (P=0.0075) and panoramic radiography (P=0.0034), but not signi®cantly dierent from D-speed ®lm (P=0.2337). Conclusions: Multidirectional tomography and panoramic radiography performed as well as intra-oral D-speed ®lm for the combined assessment of proximal and occlusal caries in the model used. When proximal surfaces were evaluated alone, D-speed ®lm was signi®cantly better. For occlusal caries there was no statistically signi®cant dierence between multidirectional tomography and D-speed ®lm. Keywords: radiography, dental; dental caries; tooth, deciduous; tomography, x-ray
Introduction Intra-oral bitewing ®lms used for caries detection have several disadvantages such as patient discomfort and variable levels of skill shown by auxiliaries. Patient radiation dose is often increased because of the retakes required. A study of intra-oral ®lms submitted to insurance companies showed that ®fty per cent of the ®lms submitted were unacceptable.1 There is no reason to assume this has changed. Compared with extra-oral radiographic techniques, the degree of co-operation and tolerance required from patients for bitewing ®lm placement is not always Correspondence to: DA Tyndall, Diagnostic Sciences Department, University of North Carolina School of Dentistry, Chapel Hill, North Carolina 27599 ± 7450, USA Received 18 July 1997; accepted 3 February 1998
attainable. It is not uncommon for pediatric and handicapped patients to require modi®cation of the standard intra-oral techniques. Many of these patients are treated without the bene®t of radiographic diagnosis.2 They would therefore bene®t greatly from an extra-oral technique that is both sensitive and speci®c for the detection of dental caries. In addition, adult patients with an increased gag re¯ex or other contra-indications to intra-oral imaging would bene®t from an extra-oral approach. Intra-oral bitewing radiographs are currently accepted as the gold standard for caries detection. The sensitivity has been reported to range from 40 ± 60%.3.4 Furthermore, as the prevalence of caries decreases, the false positive fraction increases. The characteristic appearance of proximal caries has also shifted with
Extraoral caries imaging TL Clifton et al
194
increasing exposure to ¯uoride. Caries may appear radiographically in the dentin prior to clinical cavitation of the surface of the enamel.5 Four radiographic modalities were evaluated and compared in this study. Standard bitewing D-speed ®lms were used to represent the current standard of care because this modality is still the most commonly used in private practice. Panoramic ®lms were examined since despite their low diagnostic ecacy,5 they are the most frequently used extra-oral ®lms in private practice.5 Multidirectional thin layer tomograms were evaluated because they are relatively simple to obtain with the appropriate equipment and are potentially reproducible. Initial studies in Scandinavia have shown promise for this technique in the diagnosis of periodontal and periapical lesions, but its usefulness in caries detection has not been tested.6 Linear zonograms were studied because they can be produced with fairly simple mechanics, have a wider and technically more forgiving image layer, and when used with a cephalostat are potentially reproducible. Importantly, radiation doses with linear zonography are relatively low.7 Methods The validity of D-speed ®lms, panoramic radiographs, linear zonograms and multidirectional tomograms for caries detection was determined utilizing a gold standard that had been con®rmed histologically.8 The sections were approximately 200 m thick and were oriented mesiodistally. On the basis of radiographic screening with plain ®lms and examination with a dissecting microscope, 64 primary molar teeth were selected to provide a variety of lesions as well as cariesfree surfaces. The teeth were placed in articulators with impression putty and plaster to simulate eight posterior segments with eight teeth each. All images were obtained using a 12.5 mm plexiglas block as a scattering agent. The models were also arranged with a tissue-equivalent sectioned half skull to simulate tissue absortion and artifacts encountered with crossarch extra-oral imaging techniques. Both proximal and occlusal surfaces were evaluated on all teeth. This resulted in a total of 192 surfaces for evaluation. Conventional D-speed (Eastman Kodak, Rochester, NY USA) bitewings were obtained using a object-tosource distance of approximately 40 cm. The X-ray source was a Gendex 1000 (General Electric Co., Milwaukee, WI USA) variable potential unit, at 70 kVp, 15 mA and 0.33 s. Panoramic images were produced with an Orthophos (Sirona, Bensheim, Germany) using the `Program-1' setting, and T-MAT G ®lms with Lanex regular intensifying screens (Eastman Kodak). A tube at 64 kVp and 16 mA were used. Linear zonograms were obtained with the Sectograph (American Dental, Anaheim, CA, USA) using the `wide' image layer setting for the tomographic arc and a source-to-midsagittal distance of 152 cm at 62 kVp and 50 mAs. The ®lm cassette was
placed as close as possible to the model to minimize magni®cation. The tomographic arc produced an image layer of approximately 4 mm. Multidirectional tomography was performed with the CommCAT (Imaging Sciences International, Robelin, NJ USA) multidirectional imaging system with a hypocycloidial motion producing a 6 mm thick image layer at 5 mA and 50 kVp with T-MAT G ®lm and Lanex regular screens (Eastman Kodak). After radiographs were obtained for each set of teeth, ground truth was determined by evaluating the presence of carious lesions using light microscopy as described by White and Dove.9,10 Eight dentists with experience in caries detection were instructed as to the nature of the experiment and trained in the interpretation of all imagery modalities involved. Each observer was required to express subjective certainty of the existence of caries on each surface. On a 5 point scale with 1 de®ned as certainty of absence of caries and 5 as certainty of presence of caries. The viewing sequence for each observer was a unique precomputed, balanced, and randomized distribution. The sequence was con®gured so that on average no modality was seen earlier or more frequently than any other, and so that no two sequences were alike. This assured that any learning eects were uniformly distributed among all modalities, specimens, and observers. To test the intraexaminer reliability, each observer was asked to read one modality a second time and intraclass correlation coecients were calculated from a one-way random eects ANOVA. Receiver operating characteristic (ROC) analysis was used to evaluate accuracy of caries detection.11,12 ROC curves were generated using the maximum likelihood technique in ROCFIT software (Apple Macintosh version, Charles E Metz, The University of Chicago, Chicago, IL USA). The areas under the ROC curves (AZ) were then evaluated for dierences between imaging modalities using ANOVA. Main eects and interactions of the independent variables of observer and modality were tested. The level of signi®cance was set at P=0.05. Results The distribution of histologically con®rmed lesions by location and depth is presented in Table 1. Individual and mean observer AZ scores by modality were calculated for combined, proximal, and occlusal surfaces (Tables 2 ± 5). Repeated observations were performed by only ®ve of the eight observers due to time constraints and observer compliance. These observations resulted in a minimum of one reread for each modality. Paired t-test indicated no systematic bias between the repetitions (P=0.7391). The intra-class correlation co-ecient calculated from the repetitions was 0.6197 which indicates a `moderate' to `good' level of reliability. ANOVA for individual AZ scores (Table 6) revealed that signi®cant dierences occurred due to modality
Extraoral caries imaging TL Clifton et al
(P=0.0039) in lesion detection for combined surfaces. The interaction terms observer by modality and reader by surface were not statistically signi®cant (P=0.5533 and P=0.1353). A reduced model was then examined with main eects and modality by surface interaction only. There was a signi®cant dierence among observers (P=0.0061). Reader as a main eect was therefore considered in subsequent analysis. Modality made a signi®cant (P=0.0003) dierence for detection of proximal surface lesions while observer did not (P=0.4996). Both observer and modality made signi®cant (P=0.0061 and P=0.0019) dierences in occlusal lesion detection. A post hoc test using Tukey's statistic (Table 7) revealed a signi®cant reduction in detection accuracy with linear tomography compared with D-speed ®lm
for combined lesions. D-speed ®lm was signi®cantly better than all other modalities in proximal caries. Diagnostic accuracy for occlusal caries lesion was signi®cantly better for multidirectional tomography compared with linear tomography and panoramic radiography. Pooled ROC curves were computed for all observers by modality and lesion surface (Figures 2, 3 and 4). The curves for combined and proximal surfaces reveal the superior diagnostic ecacy of Dspeed ®lm. The curve for occlusal lesions illustrates that multidirectional tomography was better than the other modalities. This dierence was statistically signi®cant for multidirectional tomography compared with linear tomography and panoramic radiography only.
Table 1 Distribution of 85 lesions on 192 surfaces of 64 primary molar teeth by region and depth
Table 3 Individual and mean performance for lesion detection on proximal surfaces with four different imaging techniques
Proximal No lesion Outer enamel caries Inner enamel caries Outer dentin caries Middle dentin caries Inner dentin caries TOTAL
Occlusal
Combined
n 96 3
% 75.0 2.3
n 11 9
% 17.2 14.1
n 107 12
% 55.7 6.3
2
1.6
23
35.9
25
13.0
19
14.8
16
25.0
35
18.2
6
4.7
5
7.8
11
5.7
2
1.7
0
0.0
2
1.0
128
100
64
100
192
Observer 1 2 3 4 5 6 7 8 Mean
0.7866 0.5170 0.8303 0.7006 0.7139 0.6246 0.7569 0.6321 0.6950
Area under ROC Curve MultiLinear directional Panoramic tomography tomography radiography 0.5537 0.4862 0.8241 0.5304 0.5869 0.5407 0.5643 0.5654 0.5815
0.7117 0.5434 0.7844 0.6830 0.5540 0.5754 0.6636 0.5899 0.6381
D-speed film
1 2 3 4 5 6 7 8 Mean
0.8254 0.6685 0.8539 0.7888 0.7579 0.7955 0.8176 0.7496 0.7822
0.4299 0.7204 0.7800 0.6252 0.6493 0.6299 0.5117 0.6070 0.6192
0.6761 0.5716 0.6737 0.6645 0.4881 0.5834 0.6374 0.5649 0.6075
0.7120 0.6794 0.6813 0.6101 0.5608 0.6199 0.7352 0.5945 0.6492
100
Table 2 Individual and mean performance for lesion detection for combined proximal and occlusal surfaces with four different imaging techniques
D-speed film
Observer
Area under ROC curve MultiLinear directional Panoramic tomography tomography radiography
0.7179 0.4849 0.7787 0.6578 0.7866 0.5640 0.6204 0.4882 0.6370
Table 4 Individual and mean performance for lesion detection on occlusal surfaces with four different imaging techniques
Observer
D-speed film
1 2 3 4 5 6 7 8 Mean
0.5365 NA* 0.7301 0.5798 0.6632 0.4379 0.4716 0.4391 0.5512
Area under ROC curve MultiLinear directional Panoramic tomography tomography radiography 0.4584 0.3692 0.7010 0.4611 0.4515 0.4454 0.6706 0.3662 0.5362
0.5765 0.6298 0.6274 0.8540 0.5889 0.5236 0.5785 0.6327 0.6264
0.5175 0.3775 0.5734 0.5949 0.5342 0.4091 0.3764 0.4091 0.5252
NA*=insucient range of responses given to generate an ROC
Table 5 Average AZ scores and standard deviation for eight observers for combined, proximal and occlusal surfaces and proximal and occlusal surfaces individually with four different imaging techniques Combined surfaces s.d. Average AZ D-speed film Linear tomography Multidirectional tomography Panoramic film
0.6950 0.5815 0.6381 0.6370
0.1011 0.1025 0.0860 0.1197
Proximal surfaces Average AZ s.d. 0.7822 0.6192 0.6075 0.6492
0.1135 0.1267 0.0990 0.0905
Occlusal surfaces Average AZ s.d. 0.5512 0.5362 0.6264 0.5252
0.0574 0.1100 0.0663 0.0615
195
Extraoral caries imaging TL Clifton et al
196
Discussion For proximal and occlusal lesions combined, D-speed ®lm was signi®cantly better than linear tomography. There was no signi®cant dierence between any other modalities. The performance of multidirectional and panoramic modalities may have been in¯uenced by the
uneven distribution of lesion depth (Table 1). While dentin lesions were fairly equal in frequency (21.09% proximal and 32.81% occlusal lesions), there was a large dierence in the number of lesions con®ned to the inner layer of enamel (1.56% proximal and 35.94% occlusal). This dierence may have resulted in a better performance of multidirectional tomography in the
Table 6 Analysis of variance of ROC curve areas (AZ) for caries detection. The dependent variable in each case is AZ. Data consists of AZ scores for eight observers and four imaging modalities. Data is presented for combined, proximal and occlusal surfaces and proximal and occlusal surfaces separately Combined proximal and occlusal surfaces: Full model Multiple R: 0.912 Multiple R: 0.831 Analysis of variance Source Sum-of-squares DF Mean square Observer Modality Surface Reader* Modality Reader* Surface modality* Surface error
0.1450 0.1066 0.3531 0.1271 0.0790 0.1576 0.1937
7 3 1 21 7 3 30
0.0207 0.0355 0.3531 0.0061 0.0113 0.0525 0.0065
Combined proximal and occlusal surfaces: Slightly reduced model Multiple R: 0.814 Multiple R: 0.663 Analysis of variance Source Sum-of-squares DF Mean-square Observer Modality Surface Modality* Surface error Proximal surfaces Source Observer Modality Error Occlusal surfaces Source Observer Modality Error
0.1497 0.1420 0.3329 0.1287 0.3868 Multiple R: 0.738 Sum-of-squares 0.0387 0.1529 0.1543 Multiple R; 0.802 Sum-of-squares 0.1770 0.1328 0.1665
7 3 1 3 58
0.0214 0.0473 0.3329 0.0429 0.0067
Multiple R: 0.544 Analysis of variance DF Mean-square 7 3 26
0.0055 0.0510 0.0059
Multiple R: 0.642 Analysis of variance DF Mean square 7 3 25
0.0253 0.0443 0.0067
F-ratio
P
3.2083 5.5034 54.7057 0.9377 1.7486 8.1402
0.0117 0.0039 0.0000 0.5533 0.1353 0.0004
F-ratio
P
3.2058 7.0987 49.9177 6.4348
0.0061 0.0004 0.0000 0.0008
F-ratio
P
0.9310 8.5924
0.4996 0.0004
F-ratio
P
3.7963 6.6434
0.0061 0.0019
Table 7 P values from Tukey's pairwise comparisons of modalities for combined, proximal and occlusal surfaces and proximal and occlusal separately for the four imaging modalities Bitewing Combined
Proximal
Occlusal
Bitewing L-tomography M-tomography Panoramic Bitewing L-tomography M-tomography Panoramic Bitewing L-tomography M-tomography Panoramic
1.0000 0.0039* 0.2264 0.2155 1.0000 0.0007* 0.0010* 0.0121* 0.4653 0.2337 0.3096
L-Tomography
M-Tomography
Panoramic
1.0000 0.2311 0.2446
1.0000 1.0000
1.0000
1.0000 0.9990 0.5862
1.0000 0.6734
1.0000
1.0000 0.0075* 0.9926
1.0000 0.0034*
1.0000
L-tomography: linear tomography. M-tomography: multidirectional tomography. P50.05
Extraoral caries imaging TL Clifton et al
combined lesion evaluation since it was better than all others for occlusal caries. When proximal surfaces were evaluated alone, D-speed ®lm performed signi®cantly better than all other modalities. This result is consistent with the literature comparing panoramic and intra-oral techniques.13
D-speed ®lm did not perform better than any other modality for occlusal caries. This is not surprising in view of the inherent diculty in identifying such lesions against an anatomically `noisy' background. The performance of multidirectional tomography was as good as D-speed ®lm and signi®cantly better than the other two modalities. The dierence in resolution between direct ®lm and screen-®lm techniques seems
a
b
c Figure 2 Averaged ROC curves for detection of combined proximal and occlusal caries by eight observers. L-Tomo=linear tomography, M-Tomo=multidirectional tomography
d
Figure 1 Typical radiographs obtained with (a) D-speed ®lm, (b) linear tomography, (c) multidirectional tomography and (d) panoramic radiography of the same teeth
Figure 3 Averaged ROC curves for detection of proximal caries by eight observers. L-Tomo=linear tomography, M-Tomo=multidirectional tomography
197
Extraoral caries imaging TL Clifton et al
198
Figure 4 Averaged ROC curves for detection of occlusal caries by eight observers. L-Tomo=linear tomography, M-Tomo= multidirectional tomography
to make little dierence in diagnosis. The greater ®lm contrast with screen-®lm combinations combined with the selective image layer of multidirectional tomography appears to improve diagnostic ecacy. The dierence with panoramic radiography, linear tomography, and intra-oral radiography in comparison with multidirectional tomography may be due to the reduction in artifact seen with wide angle (thin slice) complex motion tomography. Panoramic imaging incorporates a fairly wide image layer (15 ± 30 mm)
and thus superimposes the coronal anatomy on the region of diagnostic interest. Linear tomography narrows the image layer but introduces linear streaking. Multidirectional tomography reduces streaking and makes any artifact which is present uniform in all directions. Also, the narrow image layer reduces the contribution of unwanted adjacent anatomy and thus theoretically enhances the potential for detecting caries developing in pits and grooves. Multidirectional tomography might therefore be an alternative to Dspeed ®lm for the limited task of occlusal lesion detection with those patients where conventional bitewing ®lms cannot be used. In addition multidirectional tomography may be of some value as a means of evaluating occlusal caries in some types of patients. The future development of this, or related modalities, may result in an alternative imaging system for use in situations where an extra-oral imaging technique is the only possible method. The performance of multidirectional tomography may also be due to the ability of the tomographic system to identify the center of the image layer with a laser thus resulting in more accurate placement of the model in the image layer center. We conclude that multidirectional and linear tomography and panoramic radiography performed less well than bitewing radiography for assessment of proximal caries in the model used. For occlusal caries multidirectional tomography performed as well as bitewing radiography. Therefore for those special cases where intra-oral radiography is not possible multidirectional tomography may be appropriate. Acknowledgements The author would like to acknowledge Dr Ceib Phillips of the Department of Orthodontics for her assistance with the statistical analysis.
References 1. Beideman RW, Johnson ON, Alcoy RW. A study to develop a rating system and evaluate dental radiography submitted to a third party carrier. J Am Dent Assoc 1976; 93: 1010 ± 1013. 2. Casamassimo PS. Radiographic considerations for special patients ± modi®cations, adjuncts, and alternatives. Ped Dent 1981; 3: 448 ± 454. 3. Espelid I. Radiographic diagnosis and treatment decision on approximal caries. Community Dent Oral Epidemiol 1986; 12: 265 ± 270. 4. Douglas CW, Valachovic RW, Wizesinha A, Chauncey HH, Kapur KK, McNeil BJ. Clinical ecacy of dental radiography in the detection of dental caries and periodontal diseases. Oral Surg Oral Med Oral Pathol 1986; 62: 330 ± 339. 5. Bader JD, Shugars DA. Need for change in standards of caries diagnosis ± epidemiology and health services research perspective. J Dent Edu 1993; 57: 415 ± 421. 6. Tammisalo T, Luostarinen T, Vahatalo K, Tammisalo EH. Detailed rotational narrow beam radiography versus intra-oral radiography in detection of periodontal lesions. Dentomaxillofac Radial 1991; 20: 50 (abstr). 7. Clark DE, Danforth, RA, Barnes, RW, Burtch ML. Radiation absorbed from dental implant radiography: A comparison of linear tomography, CT scan, panoramic and intra-oral techniques. J Oral Implantol 1990; 16: 156 ± 164.
8. Pitts NB, Renson CE. Image analysis of bitewing radiographs: a histologically validated comparison with visual assessments of radiolucency depth in enamel. Br Dent J 1986; 160: 205 ± 209. 9. White SC, Hollender L, Gratt BM. Comparison of xeroradiographs and ®lm for detection of proximal surface caries. J Am Dent Assoc 1984; 108: 755 ± 759. 10. Dove SB, McDavid WD. A comparison of conventional intraoral radiography and computer imaging techniques for the detection of proximal surface dental caries. Dentomaxillofac Radiol 1992; 21: 127 ± 134. 11. Hanley JA, McNeil BJ. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Diag Radiol 1982; 143: 29 ± 36. 12. Metz CE. Basic principles of ROC analysis. Seminars in Nuclear Medicine 1978; 8: 283 ± 298. 13. Kantor ML, Zeichner SJ, Valachovic RW, Reiskin AB. Ecacy of dental radiolographic practices: options for image receptors, examination selection, and patient selection. Am Dent Assoc J 1989; 119: 259 ± 268.
