Abstract
To test the reproducibility and validity of angular measurements from virtual lateral cephalometric radiography (LCR) reconstructions (full and hemifacial on both sides) derived from cone beam computed tomography (CBCT).
Fifty-eight CBCT and LCR images were selected. CBCT volume data were imported into Nemotec software, and virtual LCR images and tomographic slices from CBCT images were assessed. Three observers digitized landmarks used for 12 angular measurements in CBCT images from all patients at two different times. The LCR were evaluated by two examiners, and the mean of the measurements was used as the gold standard.
The virtual hemifacial LCR of tomographic slices on the right side showed statistically different higher measures. The same modality on both sides showed less absolute differences for most measures except FMIA and interincisal angles. All modalities had excellent reproducibility.
The angular measurements made on virtual LCR reconstructions derived from CBCT are reproducible and valid. Some advantage was found over virtual hemifacial LCR on both sides of the image, mainly in its handling facility.
INTRODUCTION
Dentomaxillofacial radiographic diagnosis is the principal complement to clinical diagnosis of orthodontic patients. Lateral cephalometric radiography (LCR) has been used since 19311 as a standard tool for craniofacial morphology analysis, orthodontic diagnosis, and treatment.2–4
Using LCR, longitudinal studies of craniofacial growth and development were performed for analysis of dentoskeletal relationships in individuals with normal facial pattern. From these studies, standard measures of craniofacial relationships were obtained.5,6 However, LCR presents projection errors, mainly due to magnification. These errors do not affect angular measurements because of proportional magnification.7–10 Therefore, the angular measurements in cephalometric analysis on LCR images can be considered as the gold standard.1,5–7,11–13
The use of LCR in patients with severe facial asymmetry14 and in patients undergoing orthognathic surgery has decreased. In these cases, computed tomography (CT) is now a powerful tool. Cone beam computed tomography (CBCT) was introduced to the dental community 10 years ago and was specifically developed for evaluation of the head and neck.15,16 Cone beam technology has led to the development of a new generation of tomographic systems for the acquisition of volumetric images that, compared with multi-slice CT, has the following advantages: less cost, less dose of radiation, and better spatial resolution.17–19
Software development20–22 has allowed better manipulation of data on the computer. Thus, multiplanar primary reformatted (MPR) reconstruction, which is the first reconstruction of volume data from CBCT, can originate a secondary reconstruction with similar characteristics to conventional radiographs, called virtual lateral cephalograms; and for this reconstruction it is not necessary to irradiate the patient again.
This work considers three secondary reconstruction approaches. The ray-sum approach is an orthogonal projection with equal magnification between the beam's entrance and exit sides of the patient. The maximum intensity projection (MIP) approach is a method of evaluating each voxel in the volume data, showing only those voxels that have the maximum intensity of attenuation. Finally, the sagittal slice approach is made in a volume with a thickness of 0.3 mm.21
These approaches can help to fill the gaps in the traditional technique during the transition to tridimensional analysis.17, 22-24 To be accepted by the orthodontic community, it is necessary to evaluate how CBCT-based analyses can fit into the existing LCR databases.
The purpose of this study was to test the reproducibility and validity of angular measurements performed on virtual LCR reconstructions, full and hemifacial on both sides, using the ray-sum, MIP, and sagittal slice approaches, derived from CBCT and comparing them with conventional LCR.
MATERIALS AND METHODS
This study is based on images from the clinical radiology files of the Dental School of Piracicaba, at the University of Campinas and was approved by the Research Ethics Committee, at the same university.
CBCT images and LCR images of 58 individuals (38 female, 18 male; mean age 25 ± 9 years) were investigated. Conventional LCRs were obtained with a Tele Funk-15 (São Paulo, Brazil) (20 mA, 75 kV), and scanned on an HP Scanjet 4050 model in TIFF format at 300 dpi standard resolution. The CBCT images were obtained with an i-CAT Imaging Sciences International system (Hatfield, Pa); field of view (FOV) 23 × 17 cm, voxel 0.3 mm. Images with evidence of lost upper and lower incisors or the presence of supernumerary teeth over the roots of anterior teeth were excluded.
CBCT data were exported from Xoran Cat software in DICOM multifile format and imported into Nemotec (Madrid, Spain). The CBCT volume was reoriented at the midsagittal plane, the Frankfort plane, and the transporionic line. Virtual LCR models were then generated using the three-dimensional NemoCeph module from the entire volumetric data and MPR images. From volume data, ray-sum and MIP (full, right and left side) images were reconstructed (Figure 1) and from MPR images, tomographic slices were taken corresponding to the landmark used (Figure 1).
Three modalities from the virtual LCR images were considered for this research: virtual LCR full (LCRF), LCR full of maximum intensity (LCRFMI), and LCR hemifacial of tomographic slices (LCRHTS) on both sides (Table 1). Twenty-five anatomic landmarks, provided by 12 angular measurements commonly used in lateral cephalometric analysis (Figure 2), were identified on both conventional and virtual LCR images using the two-dimensional NemoCeph module. Flat-panel color active matrix TFT screens (model 19-in, AOC F19L; Envision, Sao, Paulo, Brazil; resolution 1366 × 768 at 60 Hz, 264-mm dot pitch, operated at 32-bit color) were used for the evaluations. The virtual LCR modalities had images overlapping each other, and the observer chose any image according to personal criteria based on the best landmark visualization.
Angular measurements: Tweed triangle (FMIA, FMA, IMPA), SNA, SNB, interincisal angle (U1|L1), facial depth angle (FDA), lower facial height angle (LFHA), facial axis angle (FAA), gonial angle (GA), UI inclination angle (UIIA), and facial taper angle (FTA).
Angular measurements: Tweed triangle (FMIA, FMA, IMPA), SNA, SNB, interincisal angle (U1|L1), facial depth angle (FDA), lower facial height angle (LFHA), facial axis angle (FAA), gonial angle (GA), UI inclination angle (UIIA), and facial taper angle (FTA).
Five observers previously calibrated and blinded to the identity of the images participated in this study: two for the conventional LCR images and three for the virtual LCR images. The evaluation was conducted independently and randomly at two different times with at least 2 weeks in between.
In the LCR images, the mean of the measurements of both observers was used as the gold standard. In the three virtual LCR images, two values were used. The first was the mean of the two assessments for each observer and the second was the mean of all observers, which was used for comparison with the LCR.
Statistical Analysis
The statistical analyses were performed using SPSS (version 17.0, SPSS, Chicago, Ill), and SAS (version 9.2, SAS Institute Inc, Cary, NC). The intraobserver, interobserver reproducibility and correlation between the virtual LCRs and gold standard were assessed using the intraclass correlation coefficient test (ICC) with 95% confidence levels. Analysis of variance (ANOVA) for repeated measures with the Dunnett test (P ≤ .05), absolute error (AE), and standard deviations (SD) were also calculated.
RESULTS
The ICC for the intraobserver reproducibility in all measurements ranged from 0.945 to 0.988 in the LCRF modality (mean 0.963 ± 0.010), from 0.935 to 0.985 in the LCRFMI modality (mean 0.961 ± 0.013), and from 0.965 to 0.992 in the LCRHTS on the right side (mean 0.981 ± 0.007) and from 0.969 to 0.989 on the left side (mean 0.978 ± 0.006). The ICC for the interobserver reproducibility for all angular measures ranged from 0.936 to 0.992 in the LCRF modality (mean 0.970 ± 0.015), from 0.971 to 1.000 in the LCRFMI modality (mean 0.988 ± 0.011), and from 0.984 to 0.994 in the LCRHTS modality on the right side (mean 0.98 ± 0.003), and from 0.981 to 0.993 on the left side (mean 0.988 ± 0.004).
The correlation between the virtual LCR images and the gold standard was excellent for all measurements. The ICC with the LCRF modality ranged from 0.955 to 0.982 (mean 0.973 ± 0.009), from 0.961 to 0.995 (mean 0.982 ± 0.009) with LCRFMI modality, and from 0.942 to 0.986 (mean 0.968 ± 0.010) with LCRHTS modality on the right side and 0.937 to 0.988 (mean 0.967 ± 0.012) for the LCRHTS modality on the left side.
Table 2 shows the mean values and SD for conventional LCR images and the virtual LCR images. The Dunnett test showed a statistically significant difference (P ≤ .05) between conventional LCR and the virtual LCR images for all angular measurements with some exceptions in each modality: SNB, LFHA, FAA, and FTA for the LCRF modality; SNA, SNB, FAA, UIIA, and FTA for the LCRFMI modality; FAA, IMPA, and UIIA for the right side and SNA, SNB, FAA, IMPA, and UIIA for the left side for LCRHTS modality. Table 3 shows the validity of all angular measurements determined by the AE and SD for the virtual LCR images compared with gold standard measurements.
Descriptive Statistics. Mean Measurements and Standard Deviations (SD) on the LCR (Gold Standard) and Virtual LCRs, According to the Angular Measurements (in Degrees)a

Absolute Errors (AE) and Standard Deviations (SD) of the Average Measurements of Virtual LCRs in Relation to the LCR (Gold Standard), According to the Angular Measurements (in Degrees)a

From Tables 2 and 3, we can infer for the virtual LCRF modality with statistically significant measurements, only 37.5% (3 of 8) present an AE greater than 2°, 50% (4 of 8) present an AE greater than 1°, and 12.5% (1 of 8) present AE less than 1°. On the other hand, for the virtual LCRFMI modality with statistically significant measurements, 42.9% (3 of 7) present an AE greater than 2°, and 57.1% (4 of 7) present an AE greater than 1°. Finally, for the virtual LCRHTS modality on the right side, just 11.1% (1 of 9) present an AE greater than 3°, another 11.1% (1 of 9) present an AE greater than 2°, 66.7% (6 of 9) present an AE greater than 1°, and 11.1% (1 of 9) present an AE less than 1°; for the left side 28.6% (2 of 7) present an AE greater than 3°, 57.1% (4 of 7) present an AE greater than 1°, and 14.3% (1 of 7) present an AE less than 1°.
DISCUSSION
The diagnostic LCR information is considered valuable for predicting skeletal growth and for evaluating the results of orthodontic treatment. The introduction of CBCT has provided better opportunity for orthodontic evaluation.
After the execution of computational algorithms, the CBCT data allow the generation of an infinite number of images and orthogonal cephalograms. In addition, it is possible to represent separately the right and left sides of the head without overlapping structures bilaterally. It is important to quantify the differences in the measurements, especially in individuals with severe facial asymmetry.21
The MIP approach is useful for displaying three-dimensional structures in a solid-like way and has the drawback that dense structures tend to occlude less dense structures. This might prejudice the recognition of specific anatomic landmarks (eg, sella, pterygoid, posterior nasal spine, basion, incisal edge of the lower incisor, and the apex of upper and lower incisors). The MIP images require preliminary preparation to localize these landmarks.22 In the current study, it was not necessary to manipulate the MIP images because they were combined with the ray-sum images in all the modalities evaluated.
For this research, landmarks located in the horizontal, vertical, and oblique planes were used. These planes have a proportional magnification. Therefore, the angular measurements are not influenced by the magnification minimizing the projection errors associated with head rotation on the vertical axis.5–8,13,16
There are landmarks formed by two-dimensional representation of the patient's head (eg, mandibular symphysis, pterygomaxillary fossa, articulare, key ridge).9 Nowadays, it is hard to know how the loss of information affects a cephalometric analysis. Therefore, to avoid losing this information, this study used projections with similar appearance to an LCR; new longitudinal studies are being designed to validate CBCT landmarks.
The results of the study show that the angular measurements of virtual LCR images were statistically different compared with the gold standard. For the LCRHTS modality on the right side, 75% of the measurements were different, and for the left side, using the same modality, 58% were different. For the LCRF modality, 67% of measurements were different. For the LCRFMI modality, 58% of measurements were different. Our results are in agreement with Chung et al.23 However, considering the AE, the statistical difference found probably not does reflect a relevant difference for clinical diagnostics. Our clinical results confirm and complement Kumar et al.20 who reported that the landmarks used in the conventional LCR may also be identified in reconstructed images from CBCT.
The two-dimensional projection of the patient's head in the LCR may lead to difficulty in the analysis of some structures, especially when related to dental points (incisal edge and apex points of the upper and lower incisors).11 For the incisal edge point, it is difficult to distinguish between the central and lateral incisors on the LCR. The most prominent incisor image is used to mark this point, so an error in the demarcation of this point can be expected if the lateral incisor is more prominent than the central incisor. For the apex point, the LCR image presents a lower contrast between the root apex and the surrounding bone; therefore, the location of the apex is frequently based on general observer's knowledge of tooth length. The current study showed the localization facility of those points in the LCRHTS modality. However, all the angular measurement involving the long axis of the upper and lower central incisors, were statistically different (P ≤ .05) in at least one of the modalities tested. Thus, when considering the difficulties in the analysis of the LCR, it is important to determine the validity of the measurements made in CBCT reconstructions, especially in the LCRHTS modality, which has images of specific slices of teeth.
According to the hypothesis presented in the literature about the differences in measurements related to bilateral structures,22 the present study shows that there are statistically significant differences in all measures involving bilateral points. The measures FMIA, FDA, FMA, and GA were statistically different in all the virtual LCR modalities evaluated. These results were not expected, especially in the LCRHTS modality because it presents images of each side. One possible explanation is that the pattern of superimposing anatomy differs in the conventional and virtual LCR, which may have influenced feature recognition for the localization of the bilateral points and measurements.
Angular measurements located in the median sagittal plane did not show better validity, not even in the LCRHTS modality, which has a specific image of the median sagittal region. These results were not expected because the landmarks of the medial sagittal regions are described as more stable in the literature.8 The literature also describes a significant increase in the statistically significant differences on the use of angles dependent on four landmarks.6 This was the case in the current study. Thus, all measures with those characteristics had significant differences, except the FAA angle in all modalities.
The current study shows that measurements with significant differences in the virtual LCR modalities have an AE ranging from 0.76° to 3.74° and a standard deviation ranging from 0.53° to 2.88°. In this regard, the literature indicates that in the craniofacial region evaluated by CBCT the measurement errors and SD less than or equal to 3° or 5° and 2°, respectively, are clinically acceptable.16,18
The average AE of angular measurements in all the virtual LCR modalities was less than 1.66°. However, the AE reached values of 3.74°, 3.12°, 2.71°, and 2.41° in the LCRHTS on the left side and right side, LCRF, and LCRFMI modalities, respectively. These maximum values were observed as a common pattern with the FMIA and U1|L1 angular measurements. This analysis shows that the measurements performed were independent of the virtual LCR modality used.
One limitation that was found in the ray-sum and MIP images was the lack or loss of continuity of the image of thin structures (eg, the anterior wall of the maxillary sinus, the bone overlying the teeth, and the cortical bone of the mandibular condyle); this is explained by the size of the FOV and voxels, which may have affected the location of the anterior and posterior nasal spine, porion, and condylion points. To avoid this, it is recommended that an expanded FOV is used, depending on the patient's head size, and according to the performance of the cephalometric analysis, the voxels should be as small as possible.
CONCLUSIONS
Angular measurements made on full and hemifacial LCRs on both sides derived from CBCT reconstructions are reproducible and valid when compared with measurements obtained using LCR.
Hemifacial virtual LCRs showed benefit, which could vary depending on the individual's facial asymmetry.
Hemifacial reconstruction derived from CBCT may be an appropriate alternative to replace traditional radiography in orthodontic or orthopedic diagnoses and orthognathic surgery.