The first purpose of this study was to analyze the correlation between bone volume fraction (BV/TV) and calibrated radiographic bone density Hounsfield units (HU) in human jaws, derived from micro-CT and multislice computed tomography (MSCT), respectively. The second aim was to assess the accuracy of cone beam computed tomography (CBCT) in evaluating trabecular bone density and microstructure using MSCT and micro-CT, respectively, as reference gold standards.
Material and methods
Twenty partially edentulous human mandibular cadavers were scanned by three types of CT modalities: MSCT (Philips, Best, the Netherlands), CBCT (3D Accuitomo 170, J Morita, Kyoto, Japan), and micro-CT (SkyScan 1173, Kontich, Belgium). Image analysis was performed using Amira (v4.1, Visage Imaging Inc., Carlsbad, CA, USA), 3Diagnosis (v5.3.1, 3diemme, Cantu, Italy), Geomagic (studio® 2012, Morrisville, NC, USA), and CTAn (v1.11, SkyScan). MSCT, CBCT, and micro-CT scans of each mandible were matched to select the exact region of interest (ROI). MSCT HU, micro-CT BV/TV, and CBCT gray value and bone volume fraction of each ROI were derived. Statistical analysis was performed to assess the correlations between corresponding measurement parameters.
Strong correlations were observed between CBCT and MSCT density (r = 0.89) and between CBCT and micro-CT BV/TV measurements (r = 0.82). Excellent correlation was observed between MSCT HU and micro-CT BV/TV (r = 0.91). However, significant differences were found between all comparisons pairs (P <0.001) except for mean measurement between CBCT BV/TV and micro-CT BV/TV (P =0.147).
An excellent correlation exists between bone volume fraction and bone density as assessed on micro-CT and MSCT, respectively. This suggests that bone density measurements could be used to estimate bone microstructural parameters. A strong correlation also was found between CBCT gray values and BV/TV and their gold standards, suggesting the potential of this modality in bone quality assessment at implant site.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
Primary implant stability is the key factor for the long-term success of an implant treatment by improving osseointegration (Fuh et al. 2010). Primary instability of an implant induces movements during healing. This micromotion leads to fibroplasia as a biological response at bone tissue surrounding the implant. The replacement of bone by fibrous tissue and loss of osseointegration cause implant failure (Lioubavina-Hack et al. 2006). Bone quality, which refers to the combination of all bone characteristics that influence bone resistance to fracture (Fyhrie 2005), is one of the most important factors influencing primary implant stability (Ozan et al. 2007; Tolstunov 2007). Among the bone characteristics, bone mineral density (BMD) and trabecular microstructure are the strongest predictors for bone strength (Muller 2003). However, these two parameters need to be simultaneously assessed to provide better estimation of bone strength (Teo et al. 2007; Diederichs et al. 2009). Several radiographic modalities have been used for bone quality assessment. For bone microstructure, micro-computed tomography (micro-CT) was recommended as gold standard for assessing bone morphology and micro-architecture (Burghardt et al. 2011; Ibrahim et al. 2013a). However, it is limited to ex vivo small bone samples and cannot be employed for patients. Multiple X-ray projections with different angles in micro-CT allow a precise three-dimensional (3D) reconstruction of the bone samples and assessment of bone trabeculae (Martin-Badosa et al. 2003). Micro-CT is used to measure several histomorphometric variables including bone volume (BV), total volume (TV), bone volume fraction (BV/TV), trabecular thickness (Tb.Th), trabecular number (Tb.N), and trabecular separation (Tb.Sp) (Odgaard 1997).
For bone density, multislice computed tomography (MSCT) is an established clinical modality in which calibrated Hounsfield units (HU) can accurately be converted to BMD measurements (Shahlaie et al. 2003; Shapurian et al. 2006). However, higher radiation exposure risk to patients in comparison with other modalities remains a main concern for applying MSCT for assessing bone quality (Ekestubbe et al. 1992, 1993; Frederiksen et al. 1995; Dula et al. 1996). Cone beam computed tomography (CBCT), due to increased accessibility to dental practitioners, more compact equipment and reduced cost and radiation dose, has widely replaced medical CT for oral and maxillofacial imaging. Several studies reported high geometric accuracy of CBCT for linear measurement (Naitoh et al. 2004; Lou et al. 2007; Lagravère et al. 2008), while its reliability in bone quality evaluation remains controversial. Only few studies suggested that CBCT could be applied to assess trabecular bone microstructure (Liu et al. 2007; Corpas et al. 2011). Additionally, CBCT does not represent calibrated voxel gray values expressed in HU (Hua et al. 2009). Yet, many attempts have been conducted to assess the feasibility of converting CBCT gray values to actual density measurements. High correlation between HU derived from MSCT and CBCT voxel gray values has been demonstrated, hinting at the potential of CBCT in bone density assessment (Aranyarachkul et al. 2005; Lagravère et al. 2006; Naitoh et al. 2009, 2010b; Nomura et al. 2010; Parsa et al. 2012; Reeves et al. 2012; Cassetta et al. 2013). However, the excessive scattering and technology-specific artifacts produced in CBCT have been denoted as the perpetrator for the unreliable BMD measurements (Yoo & Yin 2006; Hua et al. 2009; Araki & Okano 2011; Nackaerts et al. 2011; Schulze et al. 2011).
High correlation between bone volume fraction (BV/TV) provided by micro-CT and voxel gray value from CBCT, and also between BV fraction derived from CBCT and CT numbers from MSCT has been reported (Naitoh et al. 2010a; González-García & Monje 2012). However, the relation between BV fraction and radiographic bone density in human jaws remains controversial (Stoppie et al. 2006; Aksoy et al. 2009). Therefore, the first purpose of this study was to analyze the correlation between bone volume fraction (BV/TV) and calibrated radiographic bone density (HU) in human jaws, derived from micro-CT and MSCT, respectively. The second aim was to assess the accuracy of CBCT in evaluating trabecular bone density and microstructure using MSCT and micro-CT, respectively, as reference gold standards.
Material and methods
Sample preparation and radiographic evaluation
Twenty partially edentulous human mandibular cadavers not identified by age, sex, or ethnic group were obtained from the functional anatomy department. The cadavers were sectioned at the mid-ramus level and fixed in formaldehyde (formaldehyde 74.79%, glycerol 16.7%, alcohol 8.3%, and phenol 0.21%) and stored. A declaration was obtained from the Functional Anatomy department to use this human remains material for research purposes. The restorative materials which can produce artifact such as amalgam filling and metal crowns were removed from dentitions. The mandibles were scanned by three types of CT modalities: MSCT (Philips, 120 kVp, 222 mA, 1.128 s, 0.67 mm isotropic voxel size, Best, the Netherlands), CBCT (3D Accuitomo 170, 90 kVp, 5 mA, 30.8 s, 4 × 4 cm FOV, 0.08 mm isotropic voxel size, J Morita, Kyoto, Japan), and micro-CT (SkyScan 1173, 130 kVp, 61 mA, 35 min, 35 μm isotropic voxel size, Kontich, Belgium). In MSCT scans, the occlusal plane of each mandible was set perpendicular to the floor with zero gantry tilt, whereas in CBCT scans, it was set parallel to the floor according to manufacturer's recommended protocol. The edentulous region of each mandible was located at the center of FOV in CBCT scans. Owing to the large gantry of applied micro-CT (140 mm in diameter, 200 mm in height), mandibles were not sectioned to smaller samples. To prevent the possible micromovements during the scanning due to the large size of the samples, a cylindrical shape Styrofoam was used to fix and mount the sample into the holder.
All CT data sets were converted to Digital Imaging and Communication in Medicine (DICOM3) format. As the scan orientation differs between micro-CT and the other two systems, the z-axis of CBCT and MSCT images was flipped to match that of micro-CT for further procedures. Micro-CT data sets were large in size therefore was not possible to be flipped by our workstation. Image analysis was performed using Amira (v4.1, Visage Imaging Inc., Carlsbad, CA, USA), 3Diagnosis (v5.3.1, 3diemme, Italy), Geomagic (studio® 2012, Morrisville, NC, USA), and CTAn (v1.11, SkyScan). MSCT images were imported to 3Diagnosis software. Two cylindrical shape virtual probes (with diameter and height of 0.7 and 8 mm, respectively) were inserted at the edentulous region within the cancellous bone, with 3 mm buccolingual distance between them (Fig. 1a, b). These probes were used as indicators to facilitate the selection of exact region of interest (ROI) from MSCT, CBCT, and micro-CT. For MSCT, the probes were visible in a single cross-sectional slice as the voxel size of MSCT scans was 0.67 mm, which is thick enough to allow the probes to be visible in one slice. Subsequently, a rectangular area was drawn between the two probes from the slice of interest to define the ROI for density measurements. The mean HU values from each ROI were calculated. All ROIs were totally within the cancellous bone, excluding cortical bone, inferior dental canal, and any large bone defect.
For CBCT, a volume-based 3D registration algorithm using Geomagic software was applied to transform the inserted probes from the MSCT data sets to the CBCT scans. A standard triangulation language (STL) surface file of the MSCT and CBCT scans were matched, and the probes were transferred from MSCT scans to the exact region on CBCT's (Fig. 1c). As a result, new CBCT data sets which include the probes were obtained. Using 3Diagnosis, eight consecutive slices passing through the probes were selected from each CBCT data set to calculate the mean gray values (radiological density). This is because slice thickness in CBCT is 0.08 mm which approximately amounts to 8× thinner than the equivalent slice thickness in MSCT. A rectangular region was also drawn between the two probes similar to MSCT and gray values from corresponding anatomical locations were derived.
Cone beam computed tomography radiological density of each mandible's ROI was considered as the mean of eight calculated gray values. Subsequently, the selected ROIs were saved as a bitmap (BMP) image files to allow the trabecular microstructure evaluation.
Using Amira, each micro-CT scan was cropped to have a smaller sample including the ROI. Due to large micro-CT data sets, the superimposition of CBCT and micro-CT scans was performed as follows: maximum alignment of both data sets was obtained by manually matching and superimposition of isosurfaces generated in Amira software. Subsequently, sixteen micro-CT slices (correspondence to the eight CBCT slices) were selected and saved as a 16-BMP image file (65536 gray values). Then, these bmp files were exported to CTAn software for trabecular microstructure evaluations (Fig. 2a). A rectangular ROI for trabecular was selected on each data set slice by slice (Fig. 2b). All images were thresholded using an automated histogram analysis and binarized (Fig. 2c) to allow the measurement process. On micro-CT data sets, the ROI was again verified by carefully comparing slices with CBCT's (as reference). This was performed to reduce bias which may have been introduced during the manual superimposition of the two data sets. All measurements were performed twice with 1-month interval by a trained maxillofacial radiologist.
Statistical analysis was performed using SPSS (v17.0, SPSS Inc., Chicago, IL, USA). To determine the intra-observer reliability of the radiological and microstructural density measurement, intraclass correlation coefficient (ICC) was used. The Shapiro–Wilk test was used to verify the normality of the data. Paired t-test was used to assess the mean difference between MSCT and CBCT density measurements and between CBCT BV/TV and micro-CT BV/TV, while Pearson's correlation coefficient was used to assess the linear relation between corresponding measurement parameters. Finally, a Bland–Altman plot was used to assess the accuracy of CBCT in measuring trabecular BMD and bone microstructural density by plotting the difference between the measurements of CBCT against MSCT density and micro-CT BV/TV against the means of the compared measurements.
Excellent intra-observer reliability (ICC ≥ 0.97) was revealed for repeated measurements in the three systems. Therefore, the mean of two measurements was calculated for further analysis. The mean HU of the selected ROI ranged from −60 to 507.6 (mean 222.85 & standard deviation [SD] 140.5), while CBCT gray values ranged from 161.6 to 665.6 (mean 377.49 & SD 127.4). The negative HU derived from MSCT for case 4, 16, and 20 (Table 1) may indicate fat in trabecular spaces (Parsa et al. 2012). Calculated BV/TV of the same ROI ranged from 2.24 to 75.83 (mean 32.35 & SD 18.81) for micro-CT and from 3.73 to 68.72 (mean 36.79 & SD 23.17) for CBCT (Table 1). Paired t-test showed significant differences (P <0.001) between all comparison pairs except for mean measurement between CBCT BV/TV and micro-CT BV/TV (P = 0.147). In all selected ROIs, CBCT showed a higher density than MSCT HU and a higher BV/TV than that of micro-CT. The normal distribution of measurements was confirmed by visually inspecting the histogram and the result of the Shapiro–Wilk test (P >0.05). Therefore, the use of the t-test and Bland–Altman test is justified. Strong correlations were observed between CBCT and MSCT density measurements (r = 0.89) and between CBCT and micro-CT BV/TV measurements (r = 0.82). Excellent correlation was observed between MSCT HU and micro-CT BV/TV (r = 0.91). Bland–Altman analysis showed the bias in measuring BV/TV between CBCT and micro-CT is smaller (4.44/μm) than measuring the density between CBCT and MSCT (154.65HU) (Fig. 3a, b). The 95% measurement errors are between −21.31 to 30.19 for BV/TV and 29.74–279.56 for density measurement. The differences of CBCT and micro-CT BV/TV measurements were minimal (4.44/μm), suggesting strong agreement.
Table 1. Mean results of MSCT, micro-CT, and CBCT density (gray value) and bone volume fraction (BV/TV) measurements
It has been proven that the success of an inserted implant strongly depends on the quality, beside the quantity, of the surrounded bone (Jaffin & Berman 1991; Jemt et al. 1992). In jawbones, density measurements derived from MSCT HU are highly reliable (Schwarz et al. 1987; Shapurian et al. 2006). However, bone density alone does not fully represent bone quality and should be considered together with bone micro-architecture to estimate bone strength and fracture resistance (Diederichs et al. 2009). Histomorphometrically, bone volume fraction, which is the trabecular BV per tissue volume (TV) expressed in percentage, is the most important parameter (Parfitt et al. 1987). Micro-CT is accepted as a gold standard modality for trabecular microstructure assessment, but it cannot be employed in the clinic (Burghardt et al. 2011). In this study, our aim was to investigate the possible correlation between bone quality measurements of clinically applicable scanners in comparison with micro-CT.
A study on porcine vertebral cancellous bone revealed a high correlation between HU derived from CT images and BV/TV from micro-CT and suggested the use of HU from medical CT for the prediction of micro-architecture (Teo et al. 2006). Our results support these findings that correlation between MSCT HU and Micro-CT BV/TV is high (r = 0.91). However, the mean of calculated BV/TV in mentioned study deviated from our findings in human mandibles. This could be due to different samples, ROI selections and different scanner systems. In similar studies using human zygomatic and jawbones, a high correlation was found only in female subjects (Nkenke et al. 2003; Aksoy et al. 2009). Thus, they suggested that only female trabecular BV/TV can be predicted from BMD. In contrast, another study found a high correlation between BV/TV and HU in trabecular bone surrounded by a thin layer of cortical bone regardless to gender (Stoppie et al. 2006). This study suggested that with the development of MSCT scanners and imaging software, more precise HU measurement would be achievable (Stoppie et al. 2006). Our results showed a strong correlation between BV/TV and HU in human mandibular trabecular bone, regardless to gender and thickness of surrounding cortical bone (r = 0.91). This confirms the possibility of prediction of bone volume fraction from MSCT bone density measurement. The usefulness of this prediction can be emphasized by the limitation of micro-CT in clinical settings.
Cone beam computed tomography has several advantages over MSCT in terms of more compact equipment, small footprint for the clinic, and relatively reduced scan costs. Additionally, lower radiation dose levels to the main organs of the head and neck region have been cited as one of the most important advantages of CBCT over MSCT (Kau et al. 2005; White 2008; Carrafiello et al. 2010). Due to these advantages, the use of this modality in dental implant planning is growing so fast and it is more accessible to the dental practitioners than before. Therefore, the validity of CBCT in bone quality assessment has been studied broadly. The majority of these studies have focused on the bone density measurement and found CBCT a reliable modality for bone density measurement (Aranyarachkul et al. 2005; Lagravère et al. 2006; Naitoh et al. 2009, 2010b; Nomura et al. 2010; Parsa et al. 2012; Reeves et al. 2012; Cassetta et al. 2013). The high correlation between measured CBCT gray values and CT numbers in our study (r = 0.89) may confirms the possible potential of CBCT in radiographic density measurement. However, the limit of agreement in Bland and Altman plot (Fig. 3b) is huge (29.74–279.56) with a high bias value (mean = 154.64). This indicates an unfavorable strength of agreement. Thus, although the measurements is reliable (ICC > 0.97) and validated between two compared systems (r = 0.82), the density measurement using CBCT is less accurate when compared to its gold standard system (MSCT). It should be considered that CBCT density measurement can be effected by scanning parameters and the location of the ROI within the scanner (Nackaerts et al. 2011; Parsa et al. 2013).
Using micro-CT as gold standard, the reliability of CBCT in trabecular microstructure assessment has been validated in human mandibles, but BV/TV was not among the assessed microstructural parameters (Ibrahim et al. 2013b). Our results also confirm the reliability of CBCT in trabecular microstructure assessment, based on a high correlation between BV/TV measured by CBCT and micro-CT (r = 0.82). The positive bias value (4.44/μm) in the Bland and Altman plot (Fig. 3a) indicates that BV/TV was measured smaller by CBCT. The small range between the confidence interval for the measurement differences between the two systems was small (−21.31 to 30.19) indicates a strong agreement between CBCT and micro-CT in measuring BV/TV. In present study, smallest available FOV (40 × 40 mm) and high-resolution scan mode were applied in CBCT scans to achieve the highest possible spatial resolution (0.08 mm isotropic voxel size). Therefore, using different CBCT scanning parameters, the results may differ.
It should be emphasized that the CBCT bone quality measurements in our study deviated from those of gold standards. This deviation arises from increased scattering, noise level, and artifacts specific to the scanner technology which operates at lower peak kilovoltage and tube loading setting than MSCT and micro-CT, resulting in a reduced signal-to-noise ratio (Schulze et al. 2011). A higher noise level in comparison with MSCT can cause more inconsistencies in voxel gray values (Aranyarachkul et al. 2005; Araki & Okano 2011). Additionally, as the acquired volume in CBCT is larger than collimated fan beam in MSCT, the influence of these artifacts is excessively exacerbated (Nackaerts et al. 2011; Schulze et al. 2011).
Unlike the majority of other studies on bone volume fraction, our bone samples were not harvested for micro-CT scans. As such, in our sample, the possible deviation between the planned and excised ROI, which might arise during the trepanation procedure, was eliminated (Stoppie et al. 2006). Additionally, in the present study, a fully automated and observer independent 3D matching algorithm was employed for MSCT and CBCT scans registration to ensure that all measurements are exactly from the same site up to voxel accuracy. However, due to the manual alignment of CBCT and micro-CT data sets, there is a possibility for observer error and selection of not identical regions. As micro-CT data sets are large and therefore computationally expensive, technical limitations prohibited applying the 3D registration algorithm for automated alignment. Technical advancements in the future might resolve this issue. Finally, the difference in voxel size of CBCT (0.080 mm), micro-CT (0.035 mm), and MSCT (0.67 mm) can also contribute to the observed discrepancy in calculating BV/TV and bone density. Voxel size in CBCT influences image quality among other factors including the unit itself, tube voltage, and FOV selection (Kamburoglu et al. 2011). Generally, the smaller the voxels, the higher the spatial resolution and therefore the sharper the images appear to be. However, small voxels result in decreased contrast-to-noise ratio levels and they require higher exposure dose to the patient (Davies et al. 2012). The higher the spatial resolution, the more technical demands are imposed on the imaging system as a whole and on the imaging detector in specific to attempt to suppress noise and increase signal levels. CBCT suffers from increased noise levels especially at smaller voxel sizes due to low tube voltage, cone beam divergence phenomena, and inferior detector efficiency when compared to MSCT and micro-CT (Hassan et al. 2010). However, the potential influence of varying voxel size on visibility of hard tissue structures such as bone remains largely unknown. A recent systematic review of the literature concluded that there is a systematic lack of evidence regarding the impact of varying voxel size in CBCT on diagnostic performance and that possibly different voxel sizes might lead to comparable diagnostic outcomes (Spin-Neto et al. 2013). Only one study could be identified which demonstrated a possible effect of varying voxel size on cancellous bone measurements in micro-CT (Yeni et al. 2005). However, it remains unknown whether the same applies to CBCT. In this study, a conscious effort was made to optimize image quality through selecting the scan protocols and voxels sizes as recommended by the manufacturer for the chosen FOV's. Our results are limited to one CBCT system (Accuitomo 170), and results may vary on other systems. The design specifications of different systems still vary (De Vos et al. 2009). The lack of a technical standard for the development of CBCT systems has led to a wide disparity in the physical parameters of each model. Developing such a standard for manufacturing CBCT systems may help in generalizing research findings in the future. The study was also limited as surrounding anatomical structures including the tongue and vertebra were absent. As a result, in CBCT scans, partial object artifacts resulting from structures placed outside the scan field were not simulated. It has been previously noted that artifacts resulting from partial sampling of objects outside the scan field could result in a deviation in voxel gray values with CBCT (Katsumata et al. 2009; Araki & Okano 2011). Gray values obtained from the cadaver may also deviate from the clinical situation.
In present study, restorative materials which could induce artifacts were removed from our samples. However, in normal clinical settings, presence of metallic materials in oral cavity is quite common. In CBCT scans, restorative materials with high atomic numbers harden the X-ray beam causing streak artifacts in reconstructed images (Schulze et al. 2011), while many high-density filling materials such as amalgam completely absorb the beam causing extinction artifacts rather than beam-hardening artifacts (Haramati et al. 1994). Resulted artifacts degrade the quality of images and affect the gray scales of normal anatomical structures close to foreign bodies. The severity of mentioned effect is also dependent on the energy of applied X-ray beam, density, and geometry of artifact inducing materials (Schulze et al. 2010). Therefore, further CBCT studies on assessing the effect of dental restorative materials on bone density and microstructure measurements are suggested.
In conclusion, this present study demonstrates the reliability and validity of CBCT in bone quality assessment. However, unlike the bone volume fraction measurement, the accuracy for density measurement is unfavorable. In assessing density using CBCT, the microstructural assessment (BV/TV) is therefore recommended. However, based on the inconsistencies in CBCT designs, further studies are suggested on validation of different systems.