Reliability and accuracy of linear measurements in cone-beam computed tomography using different software programs and voxel sizes.

OBJECTIVES: The aim of this study is to evaluate the accuracy of linear measurements on cone-beam computed tomography (CBCT) images using three software programs and different voxel sizes.
METHODS: Ten human mandibles with 25 silica markers were scanned for 0.250-, 0.300-, and 0.400-mm voxels in the i-CAT New Generation (Imaging Sciences International, Hatfield, PA, USA). Thirty-five linear measurements were carried out by two examiners two times on the multiplanar reconstructions in the following software programs: XoranCat version 3.1.62 (Xoran Technologies, Ann Arbor, MI, USA), RadiAnt DICOM 2.2.9 Viewer (Medixant, Poznan- Poland) and InVesalius 3.0.0 (Centro de Tecnologia da Informação Renato Archer, Campinas, SP, Brazil). The physical measurements were made by another observer two times using a digital caliper on the macerated mandibles. ANOVA test was used to compare voxels and software programs. Pearson correlation and the Bland-atman tests were used to compare physical and virtual measurements and to evaluate the accuracy of the software programs, respectively (P < 0.05).
RESULTS: There was no statistically significant difference when the measurements were compared in acquisitions with different voxel sizes analyzed in the three software programs. There was also no difference when the measurements were compared between the software programs and the digital caliper. Excellent intra- and inter-observer reliability for the markers, physical measurements, and multiplanar reconstructions were found.
CONCLUSION: Linear measurements in the XoranCat, Radiant, and InVesalius software programs are reliable and accurate compared with physical measurements. The different acquisition protocols using different voxel sizes did not influence the accuracy of linear measurements in CBCT images.

INTRODUCTION

Cone-beam computed tomography (CBCT) represents a great advancement in dentistry, enabling improved diagnosis and planning[1] with less radiation dose and artifacts than helical CT.[2345] CBCT tri-dimensional images provide reliable measurements for clinical application,[6] but these values are frequently lower than real measurements.[67] Patient's positioning,[89] thefield of view (FOV), voxels sizes,[10] soft-tissues attenuation, and artifacts[11] may influence the image quality, varying among CBCT units and imaging protocols.[12]

CBCT original images can be exported in DICOM file format, allowing the use of public domain image processing software.[13] Despite the number of software programs used for image reformatting has rapidly increased, there is a lack of information on the relative diagnostic with regard to specific tasks.[13] As these programs use different reconstruction algorithms, it is important to evaluate their reliability to guarantee the prediction error and to establish the reliable diagnosis.[14]

Considering the variety of software programs and image acquisition protocols available, this study aims to evaluate the accuracy of mandibular linear measurements in CBCT images obtained using different software programs and acquisition protocols.

METHODS

After ethical approval (CAAE: 62133916.1.0000.0104), ten macerated human mandibles were selected without distinction of ethnicity or gender. Each mandible was positioned on the acrylic table of the i-CAT Next Generation instrument (Imaging Sciences International, Hatfield, PA, USA). The median sagittal plane was positioned perpendicular to the ground with the occlusal plane parallel to the ground to reproduce the position used for clinical examinations.[14] To standardize physical and virtual measurements, hyperdense 2 mm × 2 mm cylindrical silicon-based markers with a central orifice of 0.5-mm diameter were used.[14] They were glued directly onto the selected mandibular points with ethylene-vinyl acetate polymer.[14] The mandibles were scanned using 0.250-, 0.300-, 0.400-mm voxel sizes (acquisition time: 26.9s, 8.9s, and 8.9s, respectively; FOV 13 cm × 16 cm, 120kVp, 3–8mA).

The original CBCT data were stored in DICOM format and transferred to an independent workstation with 15.6' screen (1920 × 1080 pixels resolution) (Dell, Eldorado do Sul, RS, Brazil) running the Windows XP (Microsoft, Redmond, WA, USA). The scans were independently assessed by two experienced calibrated examiners using the XoranCat 3.1.62 (Xoran Technologies, Ann Arbor, MI, USA) - proprietary software of the CBCT equipment; RadiAnt DICOM 2.2.9 Viewer (Medixant, Poznan - Poland) and InVesalius 3.0.0 (Centro de Tecnologia da Informação Renato Archer, Campinas, SP, Brazil) - software programs available for free.

Twenty-five mandibular points were selected: Coron - point localized on the superior limit of the coronoid process (bilateral); MF - the most lower and posterior point of the mental foramen (bilateral); LLco - point localized on the lateral limit of the condyle (bilateral); MLco - point localized on the medial limit of the condyle (bilateral); AC - the most upper point of the alveolar crest on the mental foramen region (bilateral); R1 - the most concave point on the anterior border of the mandibular ramus (bilateral); R2 - the point directly opposite to R1 on the posterior border of the mandibular ramus (bilateral); R3 -the deepest point on the sigmoid notch (bilateral); R4 - the point directly opposite to R3 on the inferior border of the mandibular ramus (bilateral); Go - the midpoint on the curvature of the angle of the mandible where the ramus and the body of the mandible meet (bilateral); MandF - the most lower and posterior point of the mandibular foramen (bilateral); B - point localized in the median region between the menton and the alveolar ridge in the mental region); Me - the most inferior midpoint of the chin on the outline of the mandibular symphysis; and AC. Me - point localized on the alveolar crest in the mental region.

Thirty-five linear measurements (mm) were carried out directly on the mandibles (physical measurements) and on the software programs using the different voxels (virtual measurements): R4–R4 (distance between the left and right R4); R4-Go (distance between R4 and Go bilaterally); R4.r-Go.l (distance between right R4 and left Go); R4.l-Go.r (distance between left R4 and right Go); MF-MF (distance between the left and right MF); Go-Go (distance between left and right Go); AC-AC (distance between the left and right AC); R2–R2 (distance between the left and right R2); MandF-MandF (distance between left and right MandF); R3-R3 (distance between the left and right R3); Coron-Coron (distance between the left and right Coron); MLco-MLco (distance between the left and right MLcon); Go-Coron (distance between Go and Coron bilaterally); R1-R1 (distance between the left and right R1); R1- Co (distance between R1 and condyle bilaterally); Co-Co (distance between the left and right condyle); R1–R4 (distance between R1 and R4 bilaterally); R2–R3 (distance between R2 and R3 bilaterally); R2-Go (distance between R2 and Go bilaterally); MF-AC (distance between MF and AC bilaterally); Me-B (distance between Me and B); AC. Me-Me (distance between AC. Me and Me); AC. Me- B (distance between AC. Me and B); MLco-LLco (distance between MLco and LLco bilaterally); MLco.l-LLco.r (distance between left MLco and right LLco); MLco.r-LLco.l (distance between right MLco and left LLco); LLco-LLco (distance between right and left LLco). The physical measurements (“gold standard”) were undertaken two times, with an interval of 30 days by one observer using a digital caliper (Mitutoyo® Sul Americana Ltda, Suzano, SP, Brazil) with a 0.1-mm thick edge as a reference for the markers' central orifices [Figure 1a].

Figure 1

(a) Physical measurement with a digital caliper in a macerated mandible with silica-based cylindrical hyperdense markers. (b) InVesalius 3.0.0. software interface. Linear measurements were carried out directly on the multiplanar reconstructions (coronal, sagittal, axial)

After reformatting the multiplanar planes, the mandibular points were identified, making the virtual markers of the software programs (0.5 mm) coincide with the centers of each reformatted marker.[14] The rotation tools and the coordinates for the spatial orientation of each software were used to obtain the best visualization of the center of each marker and to confirm the location of the points. The linear measurements were carried out directly on the multiplanar reconstructions [Figure 1b], by two independent observers twice, with an interval of 30 days between the measurements in each software program.

The Shapiro-Wilk and Bartlet tests were applied to evaluate the normality of the sample and homogeneity, respectively. ANOVA test compared voxels and software programs. Physical and virtual measurements were compared by using the Pearson correlation coefficient. Bland-Atman test was applied to evaluate the accuracy of the software programs. Intraobserver and interobserver agreement was assessed using Kappa statistics. The data were analyzed on Prism 5.0 software (GraphPad Software Inc., La Jolla, CA). The significance level was set at P < 0.05.

RESULTS

Kappa test showed almost perfect intra-examiner (0.839) agreement and substantial inter-examiner agreement (0.792).

There was no statistical difference when the measurements were compared in acquisitions with different voxel sizes analyzed in the three software programs, as well as when physical and virtual measurements were compared [Table 1], but a very strong correlation (CC > 0.99) was observed between them [Table 2].

Table 1

Mean values (mm) of the physical and virtual linear measurements for the different voxel sizes (0.250, 0.300, 0.400 mm) in the Radiant, InVesalius and Xoran software programs

Table 2

Coefficient of correlation between physical and virtual linear measurements in the different software programs

The Bland–Altman test [Figure 2] comparing physical measurements with those of the Radiant, InVesalius, and Xoran software programs presented bias of 0.0864 mm, 0.0541 mm, and 0.0670 mm, respectively. These biases were close to zero and not statistically significant, showing good agreement. The most accurate software was the InVesalius.

Figure 2

Dispersion graph (Bland-Altman test) between physical and virtual measurements. The solid line indicates the mean difference between the caliper and the software. Dotted lines indicate superior and inferior limits of agreement. (a) Radiant software. (b) InVesalius software. (c) Xoran software

DISCUSSION

Measurement accuracy (validity) is defined as the degree to which a measurement represents the true value of a parameter (physical versus image measurements).[14] Numerous authors have evaluated the accuracy of CBCT images, but few studies have addressed the influence of software reconstruction on diagnostic accuracy.[13]

Currently, many software programs are available, with a wide diversity of tools with a relatively simple interface,[15] helping clinicians with diagnosis and planning within the dental specialties,[16] including measuring distances for dental implants planning,[6] orthognathic surgery or orthodontic treatment.[7] These dental software packages and applications capable of DICOM display can be categorized into proprietary and third-party commercial software. Proprietary viewers are provided by the manufacturers of CBCT equipment and act as both acquisition and viewing software. Commercial third-party DICOM viewers may not be directly associated with specific hardware.[13] Some programs must be purchased, but there are some free DICOM viewers that can be downloaded from the Internet.

Among the increasing number of software packages dedicated to managing DICOM images, the present study focused on three: a proprietary manufacturer acquisition software (XoranCat) used to acquire the images and display the data in the native format, and two third-party programs available for free (RadiAnt and InVesalius). Our results indicate that the choice of the software does not have an influence on CBCT accuracy for linear measurements in multiplanar reconstructions. We found no significant differences between physical and virtual linear measurements in the three software programs. All software packages showed high accuracy rates (CC > 0.99). The mean difference was lower than 0.02 mm, and measurement errors in 3D models of up to 1 mm are clinically acceptable for diagnosis and planning purposes.[17] The measurements showed excellent intraobserver and inter-observer reliability in all software programs and acquisition protocols, confirming the hypothesis that the software programs are reliable.

The present results corroborate other studies,[711131418] which showed that virtual linear measurements were precise when compared with the physical measurements. For Ballrick et al.,[18] the absolute difference of 0.1 mm found between virtual and physical measurements does not represent a clinical significance. Baumgaertel et al.[7] observed that although both physical and virtual measurements were highly reliable, the CBCT measurements tended to slightly underestimate the anatomic truth. Lascala et al.[6] showed that the real measurement values were always higher than those of the CBCT images, but these differences were only significant for measurements of internal structures at the skull base. All these studies used only one protocol of voxel size or did not mention this parameter.

In this investigation, the most accurate software was the InVesalius. In a similar study using linear measurements on three-dimensional surface models obtained by standard preset thresholds, Poleti et al.[14] also revealed the reliability of this program, in comparison with the Dolphin imaging software, a purchased program.

We used the cylindrical hyperdense silicon-based markers previously reported[14] because they are easily identifiable and allow precise identification of the points without metallic artifacts.[14] Poleti et al.[14] associated the excellent reliability found with the characteristics of the marker. Differently, from other studies that evaluated the precision of measurements in tri-dimensional models,[1114] the measurements were performed in multiplanar reconstructions. Further, we used twenty-five linear measurements, whereas the other authors used seventeen[14] and twenty[11] measures. We used the i-CAT unit, a large-volume CBCT that enables images for many dental specialties. The chosen protocols covered the entire mandible, using 0.250-, 0.300-, and 0.400-mm voxels. The spatial resolution is lower in faster scanning times and larger voxel sizes.[18] For this reason, we opted to compare not only the software programs but also the possible interference of the voxel sizes in the linear measurements, which is unprecedented in the literature.

Periago et al.[11] showed that the voxel size might influence the measurement precision. However, we found no differences in the measurements when different voxel sizes were used, corroborating Ballrick et al.[18] who suggested that a 0.400-mm voxel was adequate for taking measurements in craniofacial structures, providing a shorter scanning time and lower radiation exposure, with a good resolution for diagnosis or planning in most cases. Nevertheless, Liedke et al.[19] investigated CBCT images of simulated external root resorption and verified that although there was no divergence in the results for different voxel sizes, the diagnosis was easier when the smallest voxel was used, being the 0.400-mm voxel not appropriate for all patients. We suggest that reduced voxel sizes must be used only in cases that demand extremely accurate images, such as root fractures and resorption and that best protocol must be chosen considering the diagnostic and treatment planning for each patient individually.

The accuracy in clinical studies may vary because several factors, such as soft tissue attenuation, metallic artifacts and patient motion[14] and this may be a limitation of the present in vitro study. However, this study has contributed to show that, among the extensive amount of software programs available, the practitioner may opt for free packages when performing linear measurements.

CONCLUSION

Based on our results, CBCT is reliable for being applied in clinical situations where linear measurements are required. Virtual measurements were similar than those of real distances. The choice of the software and voxel size protocol does not influence CBCT accuracy for linear measurements in multiplanar reconstructions.

Financial support and sponsorship

Nil.

Conflicts of interest

There are no conflicts of interest.