Evaluation of high-resolution cone beam computed tomography in the ...

Dentomaxillofacial Radiology (2009) 38, 156–162 ’ 2009 The British Institute of Radiology http://dmfr.birjournals.org

RESEARCH

Evaluation of high-resolution cone beam computed tomography in the detection of simulated interradicular bone lesions M Noujeim*, TJ Prihoda, R Langlais and P Nummikoski Department of Dental Diagnostic Science, University of Texas Health Science Center at San Antonio, USA

Objectives: The purpose of this study is to assess the accuracy of limited-volume highresolution cone beam CT (CBCT) in the detection of periodontal bone loss. Methods: 163 simulated periodontal lesions of different depths were created in dried human hemimandibles. Specimens were imaged using the intraoral paralleling technique and limitedvolume CBCT (3DX Accuitomo; Morita Co. Ltd, Kyoto, Japan). Ten viewers examined the images. Data were analysed with receiver operating characteristics (ROC) analysis. ROC curves were generated and the areas under the maximum-likelihood curves (Az) were compared. Other statistical analyses were used to detect the normality of the distribution of the results. Results: The results are reported as the individual viewer ROC curve areas for each of the two imaging modalities. In all experiments the Az area for CBCT (0.770–0.864) was larger than the Az area for periapical film (0.678–0.783); statistical tests showed a statistically significant difference between the two modalities. Conclusions: Results indicate that the CBCT technique has better accuracy and diagnostic value than periapical films in the detection of interradicular periodontal bone defects. Dentomaxillofacial Radiology (2009) 38, 156–162. doi: 10.1259/dmfr/61676894 Keywords: cone beam computed tomography; periodontology; intraoral radiography

Introduction Since the discovery of X-rays in 1895, film has been the primary medium for capturing, displaying and storing radiographic images. It is the technology that practitioners are most familiar and comfortable with, in terms of technique and interpretation. Rapid development of computer technology, improved performance, accessibility and lower cost have opened the way to digital and advanced imaging modalities where high-performance processors are needed to handle and manipulate large amounts of data. Dental radiographs are primarily used to survey dental tissues for morphological and pathological changes of diagnostic interest. However, conventional transmission radiography projects the alveolar process onto a two-dimensional (2D) film plane so that many anatomical structures may overlie lesions in the trabecular bone. This means that the differentiation between buccal and lingual alveolar bone is limited, and *Correspondence to: Marcel Noujeim, Dental Daignostic Science, 7703 Floyd Curl Drive, San Antonio, TX 78229, USA; E-mail: [email protected] Received 12 December 2008; revised 4 April 2008; accepted 7 April 2008

this makes the topography and extent of periodontal bone lesions or dehiscences impossible to evaluate with certainty.1,2 The use of CT is common in dentomaxillofacial imaging for the diagnosis and treatment planning of malformations, impacted teeth, implants, space-occupying lesions and other pathologies requiring both direct and reformatted 2D and three-dimensional (3D) imaging; however, CT is a new technology in diagnosing dental diseases.3 Periodontal disease is a chronic bacterial infection that affects the gingiva and bone supporting the teeth. Recent epidemiologic studies show that destructive periodontitis is found in about 30% of the American population. Only careful clinical examination of the patient can detect periodontal disease at an early stage. Radiographs are necessary for showing hidden anatomical structures such as the alveolar bone. They expose the degree of interdental and interradicular bone loss, root length, crown–root ratio, periodontal ligament space and any apical pathology in the tooth.4

CBVT in detection of periodontal bone loss M Noujeim et al

Early precise evaluation of the periradicular status is necessary in diagnosis, treatment and follow-up. Although radiographs are a useful diagnostic aid, the interpretation of these images may not provide accurate information for many reasons. Bender5 described the basic principles involved in the detection of bone loss in local resorptive lesions; the results indicated that because of the low mineral content of medullary bone, large resorptive lesions in this region could go undetected; furthermore, the cortices (particularly in the mandible) have a masking effect on lesions ocurring within the cancellous bone. Several other studies have reported difficulties in accurately assessing periradicular tissues, especially for radiolucent lesions (45%¡28), including interobserver (abnormality: 50%–100%, normal features: 10%–80%) and intraobserver variability.6,7 Benn8 suggests that the current measurement techniques are insufficiently sensitive to measure 1 mm of bone loss until at least 1.9 mm of bone resorption has occurred. Besides the detection ability problem, Eickholz and Hausmann9 showed that radiographic assessment using periapical radiographs tends to underestimate the amount of bone loss by 1.41¡2.58 mm. The standard technique for the initial examination of a periodontitis patient was for many years a full-mouth periodontal probing complemented by a set of fullmouth intraoral radiographs or a panoramic radiograph with a limited number of selective periapical radiographs, depending on the severity and distribution of increased probing-pocket depths, furcation involvements or various non-periodontal findings.10 Nevertheless, the problem is that the transmission radiographs are limited because they are 2D representations of the 3D alveolar bone, tooth and soft tissue. This 2D representation is strongly affected by vertical and horizontal angulation errors during film exposure.11 The collapse of the 3D anatomy into 2D space results in the superimposition of structures that can potentially obscure features of interest and decrease diagnostic sensitivity. Consequently, a radiographic tool with a 3D presentation is preferred in pre- and post-treatment assessment of periodontal defects.12 CT scanners employing cone beam geometry are becoming popular tools in modern dental practice. Some machines cover the entire maxillofacial area while others are used for imaging a much smaller region of interest, although usually with finer resolution. Unlike conventional CT scanners, which must provide contrast resolution sufficient to visualize differences in soft tissues, scanners used in dentistry are mostly used to distinguish bone from soft tissue. As a result, noise is not as important in dental CT scanners and it is possible to get by with much fewer X-ray photons. This means that much lower-powered X-ray generators can be used (dental X-ray machines similar to those used in panoramic radiography are not uncommon) and that the radiation dose to the patient required for such scanners is much lower than that used in medical CT.13

157

Because cone beam reconstruction algorithms make it possible to reconstruct an entire volumetric region, this region can be reformatted to show anatomical detail in any imaginable plane. Views not usually seen with traditional modes of dental radiography can be achieved and accurate measurements can be performed free from the usual problems of magnification and distortion. The purpose of this study was to assess the accuracy of limited volume cone beam CT in the detection and localization of periodontal bone loss by comparing its performance with the periapical film used with the paralleling technique.

Materials and methods Simulated periodontal lesions of different depths were created in 11 dried human hemimandibles. Lesions were created in the interradicular region on the mesial or distal aspect of each tooth using a Flame 1/8 AF bur (Brasseler, GA) mounted on a high-speed. Small lesions (1–3 mm) and larger lesions (3–6 mm) were created while some regions were spared (Figure 1). Each specimen was radiographed using the two techniques; additional and deeper lesions were created and the specimens were radiographed again using the same methods. Periapical films Periapical radiographs were taken with a Prostyle Intra (Planmeca Oy, Helsinki, Finland). The measured focal spot size was 0.760.7 mm; the filtration was 2.00 mm of aluminum equivalent (EquAl). The X-ray machine was set at 60 kVp, 8 mA and with an exposure time of 0.16 s, at a focus-to-film distance of 16 inches. The dental film used was Kodak Insight (F-speed) film (Eastman Kodak Company, Rochester, NY). All films were processed on the day of exposure in an A/T 2000 XR (Air Techniques Inc., NY) automatic processor under controlled darkroom conditions. Fresh RP X-Omat chemicals (Eastman Kodak Co.) were used for processing solutions. Processor quality control procedures were carried out throughout the radiography phase of the experiment. Soft tissue attenuation and

Figure 1 A specimen with a bone defect distal to the second premolar (arrow) Dentomaxillofacial Radiology


158

scatter was simulated by placing a 1 cm thick tissue equivalent plastic in front of the specimen. Cone beam CT The cone beam CT (CBCT) machine used was the 3DX Accuitomo (Morita, Kyoto, Japan). This machine has the smallest and most limited volume size (3 cm height by 4 cm diameter) with the highest resolution of all CBCT machines.14 The specimens were embedded in water and placed using a plastic platform on the headrest of the unit (Figure 2). The orientation beam was used to align the jawbone precisely parallel to the reference surface. The scans were obtained with 180 ˚ rotation in 9 s exposure, a reconstruction matrix of 3126312 pixels and a calculated slice thickness of 0.125 mm. The tube voltage was 60 kV and the tube current 5 mA. The acquired volumes were reformatted to a thickness of 1 mm and exported and uploaded into another computer for reading. A total of 163 sites were identified of which 65 sites were control or lesion-free, 50 sites presented small lesions (1–3 mm) and 48 sites presented large lesions (3– 6 mm). All sites were in the premolar or molar regions. All periapical radiographs and CBCT volumes were coded and randomized to prevent operator bias and minimize viewer recognition of repeat images of the same mandible. Ten viewers were chosen to view the images and provide the viewer data for the study, eight with advanced training in oral maxillofacial radiology (four residents and four faculty members) and two with advanced training in periodontics. The periapical radiographs were placed in radiopaque mounts and viewed on a conventional fluorescent light view box (Ada Products Inc., Milwaukee, WI) with all light masked by a radiopaque cardboard mask except for a window of the same size as a periapical film (Figure 3a). The reading was held in a quiet room and the readers had full control of the room lighting. CBCT volumes were viewed on a Sony Vaio P4 (Sony

Corporation, Tokyo) using the i-Dixel software provided with the 3DX Accuitomo unit (Figure 3b). Viewers were asked to view images representing the two modalities at distinct viewing sessions conducted by the investigator. Each reader attended an initial orientation session to become familiar with their task and with the different techniques, 56 customized score sheets, 28 for the periapical radiographs and 28 for the CBCT volumes, were provided, presenting specific examination sites ordered in the same anatomical sequence. For each area examined, the viewer was asked to rate the confidence level of the presence of periodontal lesions using the following scale: (1) definitely absent, (2) probably absent, (3) unsure, (4) probably present and (5) definitely present. Data collection Web-based receiver operating characteristics (ROC) analysis software (http://www.rad.jhmi.edu/jeng/ javarad/roc/JROCFITi.html) was used to analyse 3260 pieces of data (163 sites 6 2 modalities 6 10 readers) from the readers. Each piece of viewer data consisted of a confidence rating from 1 to 5, indicating the viewer’s perception of the presence or absence of a simulated periodontal lesion at a specified interproximal marginal bone site as imaged by one imaging modality. These data were compared with ‘‘truth’’ as determined by an established list of lesions created by the investigator. From this comparison, true positive and false positive fractions were calculated and ROC curves (Az) generated.

Results The results are reported as the individual viewer ROC curve areas obtained by the maximum-likelihood method for each of the two imaging modalities. The mean for each of the imaging modalities is also provided. The data of one representative viewer are

Figure 2 Mandible (a) embedded in water and (b) mounted on the cone beam CT machine Dentomaxillofacial Radiology


a

159

b

Figure 3 (a) Periapical film image. (b) One sagittal slice from cone beam CT. (c) Note the bone defect mesial and distal to the first molar; this defect is not obvious on the periapical image

presented in graphical form by plotting the true positive fraction versus the false positive fraction for each of the decision criteria filled with the maximum-likelihood method of curve fit. An Az curve with an area of 1.0 represents a perfect decision, whereas a diagonal line originating from coordinates (0,0) and ending at coordinates (1,1), with an area of 0.5 represents a random decision, which is equivalent to guessing. In the first experiment, the two imaging modalities were evaluated on the basis of their diagnostic accuracy in detecting the presence or absence of simulated periodontal lesions of all sizes. The experimental or disease group consisted of all 98 sites with simulated lesions (n 5 98). The control group consisted of those sites which did not contain simulated lesions (n 5 65). The diagnostic accuracy of the two image modalities ranged from a mean ROC curve area of 0.731 for the periapical technique to an area of 0.817 for the CBCT for the maximum-likelihood method of curve fit (Table 1). The generated ROC curves (Figure 4) indicate the relative diagnostic accuracy of the two imaging modalities.

In the second experiment, the diagnostic accuracy of the two image modalities for detecting only large periodontal lesions ranged from a mean ROC curve area of 0.783 for the periapical technique to an area of 0.864 for the CBCT (Table 2) and the generated respective ROC curves (Figure 5) indicate the relative diagnostic accuracy of the two imaging modalities. In the last experiment, the diagnostic accuracy of the two image modalities for detecting the presence of simulated small size periodontal lesions ranged from a mean ROC curve area of 0.678 for the periapical technique to an area of 0.77 for the CBCT (Table 3). The ROC curves generated (Figure 6) indicate the relative diagnostic accuracy of the two imaging modalities.

Table 1 Diagnostic accuracy of imaging modalities in detecting the presence of periodontal lesions of all sizes Reader

Periapical radiograph

Cone beam CT

1 2 3 4 5 6 7 8 9 10 Mean* SD

0.750 0.740 0.750 0.720 0.742 0.718 0.700 0.697 0.740 0.753 0.731 0.021

0.880 0.849 0.806 0.792 0.839 0.813 0.778 0.776 0.802 0.832 0.817 0.033

*Mean area under the ROC curves; SD, standard deviation

Figure 4 Receiver operating characteristic curves (of all readers averaged together) for lesions of all sizes and the two modalities Dentomaxillofacial Radiology


160

Table 2 Diagnostic accuracy of imaging modalities in detecting the presence of large periodontal lesions

Table 3 Diagnostic accuracy of imaging modalities in detecting the presence of small periodontal lesions

Reader


Cone beam CT

Reader


Cone beam CT

1 2 3 4 5 6 7 8 9 10 Mean* SD

0.803 0.831 0.772 0.760 0.831 0.743 0.754 0.784 0.769 0.786 0.783 0.030

0.938 0.899 0.834 0.860 0.874 0.907 0.835 0.771 0.832 0.891 0.864 0.048

1 2 3 4 5 6 7 8 9 10 Mean* SD

0.685 0.655 0.724 0.681 0.651 0.692 0.650 0.613 0.711 0.722 0.678 0.036

0.828 0.799 0.799 0.724 0.806 0.724 0.718 0.768 0.769 0.768 0.770 0.038


Areas under ROC curves (Az) were computed for each modality and each of the ten readers, then statistically analysed using ANOVA for repeated measures to test the main effect of modality and observer, and the interaction between observer and modality. The GLM (general linear model) procedure was performed to identify the difference in the results between large and small lesions (Table 4), and it was found that the difference in ROC areas for the two modalities is similar for large and small lesions (P 5 0.6997). This means that the relative performance of one of the modalities when compared with the other will remain the same, independent of the lesion size. However, CBCT was superior to the periapical modality: F(1,9) 5 85.37 with P , 0.0001. The same test showed that large lesions are detected more easily than small lesions when examined with both modalities F(1,9) 5 113.01 with P , 0.0001. The range of Az between readers was 0.697–0.750 for periapical films

Figure 5 Receiver operating characteristic curves (of all readers averaged together) for large lesions and the two modalities Dentomaxillofacial Radiology


and 0.776–0.880 for CBCT; the differences were not significant (P 5 0.3315).

Discussion In the first experiment, the two modalities were tested for their ability to detect simulated bone lesions of all sizes. CBCT had a significantly higher diagnostic accuracy compared with periapical film which had a diagnostic accuracy of Az 5 0.731. In a similar study, Ramesh et al15 reported four different tuned aperture CT (TACT) modalities and compared them with the periapical film in the detection of simulated periodontal defects. The Az value for periapical radiography in that study was 0.640.15 The mean diagnostic accuracy of the CBCT is close to the best TACT modality accuracy of 0.820 and better than the remaining three (0.740, 0.640 and 0.690). In a recent study, Mol16 obtained very similar results in comparing introral radiography with

Figure 6 Receiver operating characteristic curves (of all readers averaged together) for small lesions and the two modalities


Table 4 The general linear model procedure Source

DF Type III SS

Mean square F-value Pr . F

Reader Size Reader/size Modality Reader/ modality Size/modality

9 1 9 1 9

0.02357023 0.09870423 0.00786102 0.07456322 0.00515502

0.00261891 1.35 0.09870423 113.01 0.00087345 0.45 0.07456322 85.37 0.00057278 0.30

1

0.00030803

0.00030803

0.16

0.3315 ,0.0001 0.8751 ,0.0001 0.9583 0.6997

DF, degree of freedom; F, F-statistic; Pr, probability; SS, sum of squares

the NewTom CBCT in the detection of periodontal bone loss. He obtained a value of 0.82 in the molar area and 0.79 in the premolar area. In our present study premolars and molars were not separated, and the Az was 0.817, which is very close to the mean Az in Mol’s study. The ANOVA test proved that the difference between the two modalities was statistically significant. Therefore, the null hypothesis was rejected and the CBCT technique proved to perform better than periapical film in the detection of interradicular periodontal bone defects of all sizes. In the second experiment, the two modalities were tested for their ability to detect large periodontal bone lesions. The CBCT performance (Az 5 0.864) was still significantly better than periapical films (Az 5 0.783) in detecting large periodontal lesions, although the Az for periapical film was also increased, probably due to the size of the lesions. The lesions were prepared in the interradicular bone without touching the buccal or lingual cortical bones, but the size of the lesions was large enough to consist of more than 40% of the bone mass in the interradicular area (45%¡28).6 These results are comparable with the results cited by Pretty and Maupome´,17 where the average Az was 0.800. In their ability to detect small bone lesions, CBCT (Az 5 0.77) was again significantly better than periapical films (Az 5 0.678) in detecting the presence of simulated small periodontal lesions. The Az area scored by the periapical film is considered relatively low, partly due to the small size of the simulated lesions covered by the intact remaining crestal bone where the lesion does not represent more than one-third of the buccolingual distance.

161

The ANOVA and related test values indicate that the difference between the two modalities is statistically significant. CBCT seems to be a very promising technique in the detection and assessment of early periodontal lesions. By comparison, markedly reduced radiation exposure can be achieved with small volume CBVT. The entire scanning of the region is performed with only one rotation resulting in a reduction in effective radiation exposure, which is reported to be 0.0074 mGy per 360 ˚ scan and 0.0037 mGy when using 180 ˚ rotation.14 In comparison, the radiation exposure involved in a digital panoramic radiograph is around 0.0063 mGy13 and in a periapical radiograph around 0.0012 mGy. Since the CBCT volume will be located in one particular region, we think that this is a small price to pay for the amount of information provided. However, being a new modality, additional training for the observer is needed to master the reading and interpretation of the images. It was noted during the study that the amount of time required by the readers for the CBCT volume was slightly longer than the reading time for films; however, decision time was shorter with CBCT. It was also noted that while reading the volumes the readers preferred using the sagittal images by moving the slices from buccal to lingual and back. Another observation is that the box plot for CBCT is wider than the box plot in periapical radiography and this is likely due to the fact that CBCT is a new imaging technique. It may take some training to understand these images, mostly because they represent thin sections of anatomy that have been imaged rather than a summation of all structures between X-ray focus and film that clinicians are more familiar with. This study was completed using dry mandibles, but although there was an attempt to produce realistic images, the same study performed on actual patients may show different results due to the presence of many other modifying factors. Another limitation of this study was the artificial nature of the periodontal defect. These vertical lesions are missing the cortical outline that delineates the outside walls of the periodontal lesion, making it more difficult to see the border between the lesion and adjacent bone marrow.

References 1. Mol A. Imaging methods in periodontology. Periodontol 2000 2004; 34: 34–38. 2. Matteson SR, Deahl ST, Alder ME, Nummikoski PV. Advanced imaging methods. Crit Rev Oral Biol Med 1996; 7: 346–395. 3. Bianchi SD, Lojacono A. 2D and 3D images generated by cone beam computed tomography (CBCT) for dentomaxillofacial investigations. CARS ’98 – Proceedings of the 12th International Symposium and Exhibition: Computer Assisted Radiology and Surgery. Amsterdam: Elsevier, 1998, pp 792– 797. 4. Mengel R, Candir M, Shiratori K, Flores-de-Jacoby L. Digital volume tomography in the diagnosis of periodontal defects: an

5. 6.

7. 8.

in vitro study on native pig and human mandibles. J Periodontol 2005; 76: 665–673. Bender IB. Factors influencing the radiographic appearance of bony lesions. J Endod 1982; 8: 161–170. Stheeman SE, Mileman PA, van’t Hof M, van der Stelt PF. Room for improvement? The accuracy of dental practitioners who diagnose bony pathoses with radiographs. Oral Surg Oral Med Oral Pathol Oral Radiol Endod 1996; 81: 251–254. Lang NP, Hill RW. Radiographs in periodontics. J Clin Periodontol 1977; 4: 16–28. Benn DK. A review of the reliability of radiographic measurements in estimating alveolar bone changes. J Clin Periodontol 1990; 17: 14–21. Dentomaxillofacial Radiology


162

9. Eickholz P, Hausmann, E. Accuracy of radiographic assessment of interproximal bone loss in intrabony defects using linear measurements. Eur J Oral Sci 2000; 108: 70–73. 10. Eley BM, Cox SW. Advances in periodontal diagnosis. 1. Traditional clinical methods of diagnosis. Br Dent J 1998; 184: 12–16. 11. Jeffcoat MK, Wang IC, Reddy MS. Radiographic diagnosis in periodontics. Periodontol 2000 1995; 7: 54–68. 12. Ramesh A, Ludlow JB, Webber RL, Tyndall DA, Paquette D. Evaluation of tuned aperture computed tomography (TACT) in the localization of simulated periodontal defects. Dentomaxillofac Radiol 2001; 30: 319–324. 13. Ludlow JB, Davies-Ludlow LE, Brooks SL. Dosimetry of two extraoral direct digital imaging devices: NewTom cone beam CT

Dentomaxillofacial Radiology

14. 15.

16. 17.

and Orthophos Plus DS panoramic unit. Dentomaxillofac Radiol 2003; 32: 229–234. Danforth RA, Dus I, Mah J. 3-D volume imaging for dentistry: a new dimension. J Calif Dent Assoc 2003; 31: 817–823. Ramesh A, Ludlow JB, Webber RL, Tyndall DA, Paquette D. Evaluation of tuned-aperture computed tomography in the detection of simulated periodontal defects. Oral Surg Oral Med Oral Pathol Oral Radiol Endod 2002; 93: 341–349. Mol A. 3D imaging of periodontal bone using cone-beam CT. Poster # 0175 The IADR/AADR/CADR 83rd General Session. (March 9–12) 2005 Baltimore, MD. Pretty I, Maupome´ G. A closer look at diagnosis in clinical dental practice: Part 3. Effectiveness of radiographic diagnostic procedures. Can Dent Assoc 2004; 70: 388–394.