ACCURACY OF BIOMARKERS OBTAINED FROM CONE BEAM COMPUTED TOMOGRAPHY IN ASSESSING THE INTERNAL TRABECULAR STRUCTURE OF THE MANDIBULAR CONDYLE

Abstract

Objective

The aim of this study was to validate the ability of CBCT to measure condylar internal trabecular bone structure and bone texture parameters accurately.

Study Design

Sixteen resected condyles of individuals undergoing TMJ replacement were collected and used as a sample. These condyles were then radiographically imaged using clinically oriented dental cone beam computed tomography (CBCT) and research oriented micro-computed tomography (micro-CT). The CBCT scans were then compared to the gold standard micro-CT scans in terms of 21 bone imaging parameters. Descriptive histological investigation of the specimens was also performed.

Results

Significant correlations were found for several imaging parameters between the CBCT and micro-CT images including trabecular thickness (r=0.92), trabecular separation (r=0.78), bone volume (r=0.90), bone surface area (r=0.79), and degree of anisotropy measurements (r=0.77).

Conclusions

Measurements of trabecular thickness, trabecular separation, bone volume, bone surface area, and degree of anisotropy obtained from high resolution dental CBCT radiographs may make for suitable bone imaging biomarkers that can be utilized clinically and in future research.

1. INTRODUCTION

The temporomandibular joint (TMJ) is one of the most complex joints in the human body and allows for both hinge-like rotational and translational movements of the mandible.¹ Similar to other diarthodial (freely moving) joints of the body, the TMJ can also be affected by osteoarthritis. Temporomandibular osteoarthritis (TMJ OA) is a degenerative disease characterized by a progressive degradation of articular cartilage and a simultaneous remodeling of the underlying bone of the mandibular condyle and glenoid fossa.² These degenerative changes that lead to structural alterations in TMJ joint morphology can also result in severe pain, loss of function, altered occlusion, difficulty opening, asymmetry, crepitus, clicking, anterior open bite, and backwards rotation of the mandible.^{3, 4}

The prevalence of TMJ OA varies across different studies due to differing diagnostic criteria, but radiographic evidence of TMJ degeneration is seen in 8–16% of the population.⁵ TMJ OA can affect either one or both TMJs, and shows a strong prevalence for women in middle age.⁶ TMJ OA is a slowly progressing disease that can take years to appear clinically, and most manifestations of TMJ OA follow a natural course independent of treatment, reaching a stable, non-painful burnout endpoint phase after a period of active joint disease.⁷ However, a small group of patients (<20%) undergo rapid and widespread degenerative changes, chronic pain, occlusal changes, and impaired quality of life, and require surgical intervention.⁸ Unfortunately there is no way to predict the course or severity of the disease, and no single clinical finding specific to TMJ OA exists.⁷

The initial morphological changes in TMJ OA are subtle and subclinical, with or without significant symptoms, and therefore many patients may not present symptoms in the early stages. Unfortunately after such joint changes have occurred and TMJ OA reaches a point where it is clinically and radiographically detectable, it is not possible to restore the condyle to its original morphology.⁹

Medical CT is currently the gold standard for imaging degenerative osseous changes associated with TMJ OA and has shown an 84% accuracy in detecting these arthritic changes.¹⁰ However, medical CT has many drawbacks such as size and cost of the machine, cost per scan, high radiation dose per scan, and unfamiliarity to dentists who are often the first point of contact for patients with TMJ disorders.

Recently, CBCT has emerged as the radiographic modality of choice for imaging the osseous components of the TMJ as it has been shown to be as effective as conventional medical CT in detecting bony changes in the joint.^{11, 12}

Several studies have observed specific architectural changes in the subchondral bone of other joints that have been also affected by osteoarthritis, such as the knee, hip or vertebrae.^13–15 These findings support the notion that changes in the subchondral bone of the TMJ may be indicative of an underlying disease process, and thus could possibly be used for early diagnosis. Therefore, analysis of the internal trabecular architecture of the TMJ may potentially be a method to diagnose the disease early without having to wait until frank evidence of surface changes of cortical bone are visible much later in the disease process.

Several different structural bone parameters have been studied in the literature to assess the internal structure of trabecular bone, such as trabecular thickness, trabecular separation, bone volume, bone volume fraction, bone surface area, degree of anisotropy, structural model index, ellipsoid factor, connectivity, connectivity density, and the Euler characteristic.^{16, 17} Furthermore, another technique that has been commonly used to compare images in biomedical imaging literature is grey level run-length. A grey-level run is a set of consecutive, collinear picture points having the same grey-level value. The length of a certain run corresponds to the number of collinear picture points. The computation of grey level run-length features is based on the grey level run length matrix that describes each pixel’s neighborhood local texture. Different grey level run- length features correspond to different values for homogeneous, heterogeneous, coarse, or fine textures.¹⁸ To date, no studies have attempted to compare bone texture in CBCT and micro-CT images using grey level run-length analysis. The aim of this study is to validate the ability of CBCT to measure condylar internal trabecular bone structure and bone texture parameters accurately.

2. METHODS & MATERIALS

The protocol for the CBCT assessment of the internal trabecular structure of the mandibular condyle was reviewed by the University of Michigan IRB and was deemed to have ‘not regulated’ status and thus did not require IRB approval.

2.1 SAMPLE

A sample of 16 resected condyles was collected from 14 patients undergoing total joint replacement of either one or both TMJs. Condylar resections were performed by one operator, an oral and maxillofacial surgeon, in Dallas, Texas. A clinical examination, based on the Diagnostic Criteria for Temporomandibular Disorders (DC/TMD),¹⁹ was performed on all patients by a clinician trained and experienced in DC/TMD examination and diagnoses. All cases met the DC/TMD axis I group III criteria for TMJ OA¹⁹ and the radiographic diagnosis criteria²⁰ were assessed by a cone beam CT taken for diagnostic purposes.

All condyles in the sample were from females ranging in age from 18–64 years. The review of the patients’ medical history excluded any subjects with systemic or metabolic conditions. The samples were fixed in 10% neutral buffered formalin and then stored in 70% ethanol at freezing temperatures until analysis was performed.

2.2 3D RADIOGRAPHIC IMAGE ACQUISITION

CBCT scans of each resected condyle were obtained at the University of Michigan School of Dentistry using a 3D Accuitomo 170 (J Morita Mfg Corp., Tokyo, Japan) using a localized 40 mm × 40 mm small field of view. All samples were carefully oriented in a positioning cup to standardize and calibrate their placement within the acquired CBCT images. The primary reconstruction of the acquired volumetric images was performed in isotropic 80 µm voxels with 500 axial slices. Exposures were made using an 18 second scan time at 90 kVp and 1 mA.

Micro-CT scans of each resected condyle were also obtained at the University of Michigan School of Dentistry using a µCT 100 unit (Sanco Medical, Bssersdorf, Switzerland). Specimens were embedded in 1% agarose prior to scan acquisition. Scans were acquired at a medium resolution setting of 40 µm voxel size with an integration time of 500 ms at 90 kVp and 155 µA.

2.3 3D RADIOGRAHPIC IMAGE ANALYSIS

The image analysis was performed in 6 steps.

Step 1 - 3D reconstructions of the CBCT and micro-CT scans

Reconstructions of the 3D volumetric label maps from the CBCT and micro-CT scans were performed using ITK-SNAP v. 2.4 (open-source software, www.itksnap.org). First, DICOM files from both the CBCT and micro-CT scans were imported into ITK-SNAP. Trabecular bone was segmented and labeled using a semi-automated one-sided thresholding procedure in which voxels with grey values above a given value were labeled. For the CBCT images, this lower threshold value was initially set to a grey scale value of 1,400 and then manually adjusted to best visualize the trabeculae. Lower threshold values were maintained within a range of 1,200 to 1,600 to ensure consistency. Due to the different intensity of the micro-CT scans, the lower threshold value of these scans was initially set to 21,000 and adjusted within a range of 19,000 to 23,000 to best visualize the trabeculae. The final 3D volumetric label maps were converted and then saved as .stl files using the ‘Model Maker’ tool in 3D Slicer (open-source software, www.slicer.org).

Step 2 - Registration of the micro-CT to CBCT scans

As the orientations of the CBCT and micro-CT scans were not the same, there was a need to reorient and register the 3D surface models such that both the CBCT and micro-CT scans could be compared to one another. The 3D surface models were opened in 3D Slicer and 3D models of each condyle for both CBCT (red) and micro-CT (beige) scans were approximated and registered (Figure 1A).

Figure 1.

A) 3D surface models of the condyles from both the CBCT (red) and micro-CT (beige) scans were approximated and registered in Slicer software. B) A representative trabecular region of interest for both CBCT (left) and micro-CT (right) images. Region of interest crop boxes measured 2.48 mm³ for both imaging modalities and were registered on the same coordinate plane. C) Cropped CBCT and micro-CT 3D volume models generated from regions of subchondral bone defects.

A two-step registration process was employed to register the corresponding micro-CT and CBCT 3D surface models for each condyle. First, using the ‘Transform’ tool within 3D Slicer, the micro-CT 3D surface model for each condyle was manipulated and approximated to the corresponding CBCT 3D volume model, which served as a template. This approximation generated a transformation matrix that contained data quantifying the rotations and translations used to reach the manual approximation. The transformed 3D surface model of the micro-CT was saved.

Second, the approximated CBCT and micro-CT 3D surface models of each condyle were then registered to one another by performing a surface registration using the ‘CMF Registration’ tool in 3D Slicer. This tool automatically manipulated the micro-CT model to achieve the best fit on the CBCT model. Similarly, this step also generated a matrix that quantified the subsequent transformations needed for the registration. The registered 3D surface model of the micro-CT was saved.

Lastly, the two transformation matrices for each condyle – the approximation matrix and the registration matrix – were applied to the original micro-CT scans to reorient and register them to the corresponding CBCT scan. Then both the micro-CT scans and 3D volume models were reformatted to the same voxel size as the CBCT in 3D Slicer. The goal of normalizing the voxel size between CBCT and micro-CT was to test whether bone texture is detected similarly in both imaging modalities at the smaller voxel size currently possible for TMJ imaging protocol of patients.

Step 3 - Identification of subchondral bone defects in CBCT & micro-CT scans

In order to identify regions of interest of subchondral bone defects, the registered CBCT and micro-CT scans for each condylar specimen were opened side-by-side and scrolled through slice-by-slice. A subchondral defect was recorded as an area in the bone beneath the cortical surface that appeared as a cyst and where the regular trabecular architecture was interrupted. Subchondral cysts are described as radiographic criterion of TMD.¹⁹ Ahmad et al²⁰ explain that the term subcortical or subchondral cyst is in fact a misnomer since it is not a true cyst but rather a region of osseous degeneration. The xyz coordinates at the center of each region of interest were recorded.

Step 4 - Trabecular image cropping

At each region of interest, the CBCT and micro-CT scans and 3D volume models were cropped by 15 slices in each direction from the recorded xyz coordinates. Thus for each region of interest, square crop boxes of each volume measuring 31×31×31 slices or 2.48 mm×2.48 mm×2.48 mm were generated (Figure 1, B and C).

Step 5 - Quantification of trabecular architecture

Quantification of trabecular architecture was done using ImageJ software (open-source software, https://imagej.nih.gov/ij/index.html). The CBCT and micro-CT 3D volume crop boxes of each region of interest were opened in ITK-SNAP and exported as MetaImage files (.mha), which could then be imported into ImageJ. The 3D crop boxes were then analyzed using the BoneJ plug-in (open-source software, www.bonej.org) for trabecular thickness, trabecular separation, bone volume, bone volume/total volume, bone surface area, degree of anisotropy, structural model index, and ellipsoid factor. The 3D crop boxes were then processed using the ‘Purify’ tool in BoneJ to condense the trabecular network into a single interconnected element that could be used as an input into the connectivity, connectivity density, and Euler characteristic analyses. BoneJ has been validated as an accurate tool to measure many of these bone structure parameters.²¹

Step 6 - Bone texture analysis

Grey level run-length coding was performed on the crop boxes of the CBCT and micro-CT scans and the 10 grey level run-length bone texture parameters from Table 1 were analyzed using Linux scripts.

Table 1.

Comparison of grey level run-length bone texture parameters from the cropped CBCT and micro-CT volumes of the defective regions.

Grey Level Run-Length Parameters (Cropped Scans)	Correlation	CBCT (mean±s.d.)	Micro-CT (mean±s.d.)	Difference
Short Run Emphasis	−0.14	0.76±0.09	0.95±0.05	−0.19***
Long Run Emphasis	−0.33	262±145	27±43	235***
Grey Level Nonuniformity	−0.05	401±67	486±79	−86**
Run Length Nonuniformity	−0.43	7666±1333	12535±2229	−4869***
Low Grey Level Run Emphasis	0.03	0.08±0.02	0.10±0.02	−0.02**
High Grey Level Run Emphasis	0.73***	1257 ±2 91	1466±333	−209***
Short Run Low Grey Level Emphasis	0.04	0.05 ± 0.02	0.09±0.02	−0.03***
Short Run High Grey Level Emphasis	0.61**	966 ± 217	1372±246	−406***
Long Run Low Grey Level Emphasis	−0.05	33.53 ± 21.03	9.37±7.55	24.16***
Long Run High Grey Level Emphasis	−0.18	330462 ± 223708	57458 ± 138340	273003***

^*p < 0.05,

^**p < 0.01,

^***p < 0.001

2.4 HISTOLOGY

After radiographic image acquisition, the histologic evaluation was performed in 3 steps: the condyle specimens were sectioned, stained, and analyzed using light microscopy to assess localization of osteoclasts on the defective trabecular regions.

Step 1 - Sectioning

The condyles were decalcified with 10% EDTA for 6 weeks. After decalcification, specimens were dehydrated using a cold graded ethanol series. The 3D surface models of each specimen were used to carefully identify the direction and location for sectioning, marking, and annotating how the histology sectioning should be performed to match the scans’ slices as closely as possible. One previously identified subchondral defect was picked per specimen as the region of interest to prepare. In order to ensure that the histological sections transected the chosen subchondral defect, the distance of the defect from an easily identifiable surface (reference surface) was measured radiographically by counting how many slices the defect was from the surface and then multiplying this by the voxel size (Figure 2). For instance, if the defect was 50 slices from the reference surface and the voxel size was 80 µm, then the defect was 4 mm from the reference surface. Specimens were first embedded in paraffin wax. Sectioned cuts of 6 µm thickness were then made parallel to the reference surface at a depth corresponding to the depth of the regions of interest.

Figure 2.

Schematic picture showing the methodology that was used to determine the sectioning planes for histology and the depth of the slices. A reference plane was chosen in relation to a readily identifiable surface, the distance from the defect was measured perpendicular to this plane and the sectioning cut was then made parallel to the reference plane at the appropriate depth.

Step 2 - Staining

The biological tissues were then stained with tartrate-resistant acid phosphatase (TRAP) to visualize osteoclasts.²² Sections were washed with 0.1 M acetate buffer (pH 5.0). They were then incubated with naphthol AS-MX phosphate and red violet LB salt (Sigma, St Louis, Mo), which was diluted in 0.1 M acetate buffer (pH 5.0) containing 50 mM/L tartaric acid. Sections were incubated at 37°C for 30 minutes. Sections were then counterstained for hematoxylin (WAKO, Osaka, Japan) and mounted on slides.

Other sections were also stained with hematoxylin and eosin (HE), which is the gold standard stain to visualize bone.²² Sections were then mounted on slides.

Step 3 - Microscopy and Histomorphometry

Histomorphometric analysis was done using Nikon NIS Elements Advanced software (Nikon Instruments Inc, Tokyo Japan). The defect of interest was identified and these specific areas were photographed at 100x to visualize any TRAP positive cells. Five sequential slices for each sample were photographed at the defect location. Osteoclasts were defined as TRAP-positive multinucleated cells on the bone surface or in resorptive lacunae. Trabecular bone perimeter was also calculated using the measure distance tool in NIS-Elements by manually tracing the bone perimeter in each of the slides. Osteoclasts were counted by two independent observers for 5 sequential slides. The counts were averaged between the two observers.

2.5 STATISTICAL ANALYSIS

For each CBCT and micro-CT volume of each region of interest, 10 grey level run-length bone texture parameters (Table 1) and 11 trabecular bone structure parameters (Table 2) were computed. Paired t-tests were chosen to evaluate if there was a consistent difference in the mean values of the 21 imaging parameters as measured on the CBCT and micro-CT scans. Pearson correlation analysis was chosen to assess if the same measured variable was linearly correlated between the two different imaging modalities.

Table 2.

Comparison of trabecular bone structure parameters from the cropped CBCT and micro- CT volumes of the defective regions.

Bone Structure Parameters (Cropped 3D Volumes)	Correlation	CBCT (mean±s.d.)	Micro-CT (mean±s.d.)	Difference
Trabecular Thickness Mean (mm)	0.92***	1.12±0.55	0.67±0.34	0.44***
Trabecular Separation Mean (mm)	0.78***	1.04±0.36	0.84±0.30	0.21***
Bone Volume (mm3)	0.90***	7.78±3.07	6.69±2.61	1.09**
Bone Volume/Total Volume	0.90***	0.51±0.20	0.44±0.17	0.07**
Bone Surface Area (mm2)	0.79***	25.78±8.86	21.21±10.21	4.57**
Structural Model Index	0.60**	3.34±0.62	3.63±1.22	−0.29
Degree of Anisotropy	0.77***	0.76±0.19	0.73±0.15	0.03
Euler Characteristic	0.35	−3.30±3.08	−23.25±19.32	19.95***
Connectivity	0.47*	5.67±3.75	26.19±19.29	−20.52***
Connectivity Density (/mm3)	0.47*	0.37±0.25	1.72±1.26	−1.35***
Ellipsoid Factor (%)	0.28	11.92±6.53	12.25±4.48	−0.33

^*p < 0.05,

^**p < 0.01,

^***p < 0.001

Similarly, a Pearson correlation analysis was also carried out to determine if the osteoclast number and density from the histomorphometric analysis were linearly correlated with any of the 21 imaging parameters calculated from the CBCT and micro-CT images.

3. RESULTS

3.1 COMPARISON OF TRABECULAR BONE STRUCTURE PARAMETERS BETWEEN THE CBCT AND MICRO-CT VOLUMES

The 11 trabecular bone structure parameters that were computed from the cropped CBCT and micro-CT 3D volumes of the defective trabeculae regions were first compared using paired t-tests, which showed very similar findings (Table 2). Measurements of trabecular thickness, trabecular separation, bone volume, and bone surface area were significantly greater on the CBCT images than on the micro-CT images. The measurements of the Euler characteristic, connectivity, and connectivity density were significantly lower in the CBCT images than on the micro-CT images. Lastly, no significant difference was observed for the structural model index, degree of anisotropy, or ellipsoid factor.

The Pearson correlation coefficients showed similar trends between the trabecular bone structure parameters measured on the cropped CBCT and micro-CT volumes (Table 2, Figure 3). Strong correlations (p < 0.01) of greater than r = 0.6 were observed for 7 of the 11 trabecular bone structure parameters.

Figure 3.

Scatterplots showing correlations and linear regression between the trabecular bone structure parameters measured on the cropped CBCT and micro-CT volumes. Only statistically significant associations are illustrated. Insignificant associations were omitted. Note: * p < 0.05, ** p < 0.01, *** p < 0.001.

3.2 COMPARISON OF BONE TEXTURE PARAMETERS BETWEEN THE CBCT AND MICRO-CT VOLUMES

The paired t-tests between the 10 grey level run-length bone texture parameters measured on the cropped CBCT and micro-CT scans showed a significant difference in all bone texture measurements (Table 1). Strong Pearson correlation coefficients were observed for high grey level run emphasis and short run high grey level emphasis parameter (p < 0.01).

3.3 HISTOMORPHOMETRY OF OSTEOCLAST NUMBER & DENSITY

A summary of the bone perimeter, number of osteoclasts, and osteoclast density of each sectioned specimen can be found in Table 3. Although osteoclasts were not found in all specimens, it appeared that the osteoclasts were increased in regions close to the subchondral defects in most specimens. Pearson correlations between osteoclast number and density and the 21 imaging variables from the CBCT and micro- CT volumes for the corresponding specimens were calculated (Figure 4).

Table 3.

Summary of the bone perimeter, number of osteoclasts, and osteoclast density of each specimen.

Specimen	Bone Perimeter (mm)	# of Osteoclasts	Osteoclast Density (# of osteoclasts/mm)
1L	5.26	8	1.51
2L	6.13	9.8	1.58
2R	3.90	5.6	1.42
3L	5.44	7.3	1.41
3R	6.39	9.8	1.55
4L	3.06	4.4	1.35
5L	3.59	0.7	0.21
5R	3.82	0	0
6L	4.50	4.9	1.11
6R	3.02	3.1	1.03
7R	4.48	3.8	0.89
8L	3.65	4.8	1.35
9L	4.96	4.3	0.87
10R	3.39	5.5	1.37
11R	4.73	2	0.43
13R	5.04	10.6	2.08

Figure 4.

Scatterplots showing correlations and linear regression between osteoclast number and density and the trabecular bone structure parameters from the CBCT and micro-CT volumes for the corresponding specimens. Only statistically significant associations are illustrated. Insignificant associations were omitted. Note: * p < 0.05, ** p < 0.01, *** p < 0.001.

For CBCT (Table 4), mild positive correlations were observed between osteoclast number (0.42) and osteoclast density (0.48) with ellipsoid factor. Mild negative correlations were also observed between osteoclast number and low grey level run emphasis (-0.46) and long run low grey level emphasis (-0.44).

Table 4.

Pearson correlations of osteoclast number and density to 21 imaging parameters from CBCT and micro-CT.

Measurements (n = 26)	Correlation with CBCT		Correlation with Micro-CT
	Osteoclast #	Osteoclast # / mm	Osteoclast #	Osteoclast # / mm
Trabecular Bone Structure Parameters
Trabecular Thickness (mean) (mm)	−0.34	−0.38	−0.36	−0.36
Trabecular Separation (mean) (mm)	0.30	0.34	−0.17	−0.01
Bone Volume/Total Volume	−0.34	−0.35	−0.08	−0.12
Bone Surface Area (mm2)	−0.28	−0.32	0.14	0.07
Structural Model Index	0.03	0.02	−0.29	−0.28
Degree of Anisotropy	0.26	0.33	0.00	0.08
Euler Characteristic	0.23	0.28	−0.56**	−0.33
Connectivity	−0.14	−0.18	0.52**	0.31
Connectivity Density (/mm3)	−0.14	−0.18	0.52**	0.31
Ellipsoid Factor (%)	0.42*	0.48*	−0.32	−0.34
Grey Level Run-Length
Bone Texture Parameters
Short Run Emphasis	−0.16	−0.11	0.06	0.19
Long Run Emphasis	0.19	0.16	−0.19	−0.33
Grey Level Nonuniformity	−0.07	−0.05	0.54**	0.35
Run Length Nonuniformity	−0.36	−0.23	0.27	0.18
Low Grey Level Run Emphasis	−0.46*	−0.36	0.35	0.32
High Grey Level Run Emphasis	0.16	0.13	−0.42*	−0.39*
Short Run Low Grey Level Emphasis	−0.21	−0.13	0.30	0.33
Short Run High Grey Level Emphasis	−0.001	−0.02	−0.41*	−0.31
Long Run Low Grey Level Emphasis	−0.44*	−0.37	0.08	−0.05
Long Run High Grey Level Emphasis	0.32	0.31	−0.33	−0.43*

^*p < 0.05,

^**p < 0.01,

^***p < 0.001

For micro-CT (Table 4), moderate positive correlations were found between osteoclast number and several trabecular bone structure and bone texture measurements including: connectivity (0.52), connectivity density (0.52), and grey level nonuniformity (0.54); moderate negative correlation was found between osteoclast number and the Euler characteristic (-0.56) (p < 0.01). Mild negative correlations between osteoclast number and high grey level run emphasis (-0.42) and short run high grey level emphasis (-0.41) were observed (p < 0.05). Similarly, mild negative correlations between osteoclast density and long run high grey level emphasis (-0.39) and high grey level run emphasis (-0.43) were observed.

4. DISCUSSION

CBCT has been successfully used to characterize the osseous surface morphology of the maxillofacial region.²³ Current CBCT machines can now resolve up to 80 µm,²⁴ which theoretically should be more than sufficient to image healthy human trabeculae that range in thickness from approximately 200–400 µm.^{25, 26} However, few studies have been done to determine if this is the case and if the contemporary CBCT machines used in dental practice today can in fact be used to accurately characterize the internal structure of trabecular bone, for instance, in the mandibular condyle. Knowledge of such internal trabecular bone parameters may provide another aspect of bone morphology, which can aid in the diagnosis and monitoring of TMJ OA.

The results from our study show that the CBCT images had significantly higher measures for trabecular thickness, trabecular separation, bone volume, bone surface area and significantly lower measures for Euler characteristic, connectivity, and connectivity density as compared to the micro-CT. This was the expected result and can be attributed largely to the difference in voxel size between the CBCT and micro-CT images (80 µm vs. 40 µm). Secondly, when trabecular bone is in fact detected within a voxel in a CBCT scan (even if in reality the trabecula does not completely fill the entire area of the voxel), a single grey value will be registered for the entire voxel. In comparison, this same area would be comprised of 8 different voxels in the higher resolution micro-CT scan, and thus would allow for varying grey level values to be registered depending on the amount of bone present in each of these 8 voxels. Thus, a larger voxel size not only omits a subset of the data that is below its threshold of detection, but also creates a rounding error in them. Simply put, while fine structures are not detected on CBCT, thicker structures may appear even larger and dimensionally inaccurate. This is known as the partial volume averaging effect, which has been shown to contribute to erroneous bone dimensions and bone volume fractions in various medical imaging modalities.^{27, 28} These findings were also qualitatively observed in the visual comparison of the CBCT and micro-CT scans (Figure 1). Additionally, the image analysis procedures in the present study included multi-modality registration to compare CBCT and micro-CT scans. While Hill et al²⁹ described the registration of images taken with the same modality to assess longitudinal changes, Behnami et al³⁰ discussed the use of multi- modality registration to improve anatomic interpretation.

While the absolute values for most of the bone parameters measured were significantly different between the CBCT and micro-CT scans, high Pearson correlations were observed for many bone structure parameters between the CBCT and micro-CT scans. Specifically, the trabecular thickness, trabecular separation, bone volume, bone surface area, and degree of anisotropy showed strong significant correlations for the defect region. This indicates that these structural bone parameters may be utilized as potential bone imaging biomarkers in future CBCT radiographic studies of TMJ OA.

The only bone texture parameter that showed strong significant correlation between the CBCT and micro-CT was the high grey level run emphasis that may be used as potential imaging biomarker. All of the other grey level run-length bone texture parameters showed either no correlation or a weak correlation and, thus, none of these parameters are considered to be good candidates for possible radiographic biomarkers. Hounsfield units were not used to assess the grey level measurements in the present study. Parsa et al³¹ and Pauwels et al³² reported that grey values in CBCT scanners are higher than the Hounsfield units derived from MSCT, due to increased noise, scattering, and artifacts in CBCT imaging. Pluim et al³³ noted the challenges in using mutual information registration to measure the distribution of co-occurring grey intensity values in multi-modality comparisons. The registration procedures in this study were not used for image subtraction or measurement of grey level differences. The registration in this study simply assured that scans in both imaging modalities were assessed in similar orientations.

Our study also found osteoclasts localized in regions of subchondral defects in TMJ OA. This observation agrees with a previous study that found osteoclasts in subchondral cysts in cases of osteoarthritis of the femur.³⁴ However, to the best of our knowledge, ours is the first study attempting to correlate osteoclast number and density with radiographic imaging parameters for TMJ OA. An interesting finding of our study was the moderate correlations found between the number of osteoclasts in the histological sections of the defective regions of the specimens and various bone structure parameters measured from the micro-CT volumes – namely connectivity, connectivity density and the Euler characteristic. These findings suggest that 3D imaging can give us insight into the degenerative cellular processes occurring within the condyle. For instance, when parameters are increased, degenerative metabolic processes may be at play that up- regulate osteoclasts. However, it is difficult to directly draw this conclusion as confounding factors may have contributed to the results, and micro-CT measurements cannot be clinically applied due to their inherent limitations. For example, connectivity is an approximation of the number of trabeculae in the sample and connectivity density gives an approximation of the density of trabeculae (i.e. trabeculae per mm³), while the Euler characteristic is a mathematical representation of how many connected structures there are in a network. As can be seen, all these measures are themselves correlated to the number of trabeculae. Therefore, it may simply be that in a sample with more trabeculae, there is a greater likelihood of detecting osteoclasts as there are proportionally more regions in which they can be localized. It is also important to note that these 3 bone structure parameters only showed significant correlations to the raw number of osteoclasts per slide but not to osteoclast density. Thus, the correlations observed may be more of a result of the parameter being measured rather than an underlying biological process. Further, it was also found that osteoclast number and density were significantly correlated with ellipsoid factor measured from CBCT volumes; however, ellipsoid factor from the micro-CT volumes, the gold standard, was not significantly correlated with either osteoclast number or density and thus this is likely an aberrant finding.

There are only two other studies in the literature that have investigated the accuracy of CBCT compared to micro-CT in imaging mandibular structures.^{35, 36} Each of these studies compared CBCT and micro-CT in measuring a different set of trabecular bone parameters and our results generally agree with the findings in these studies. Ibrahim et al.³⁵ demonstrated a similar Pearson correlation between micro-CT and CBCT for the measurement of trabecular thickness in the posterior mandible of human cadavers (r = 0.82) and also showed a significant correlation for trabecular separation (r = 0.94) between the two imaging modalities. Ho et al.³⁶ demonstrated that CBCT and micro-CT measurements of trabecular thickness, trabecular separation, bone surface area, and bone volume fraction were all highly correlated with Spearman correlations ranging from 0.8 to 0.96. However, a significant difference between our study and a limitation of their study was that they imaged uniform synthetic bone blocks made from polyurethane plastic.

The results of our study were quite encouraging in that the trabecular bone structural parameters measured from volumes acquired on a CBCT machine commonly used in the dental setting are representative of the true nature of bone. Specifically, trabecular thickness, trabecular separation, bone volume, bone surface area and degree of anisotropy are structural bone parameters that show promise to be utilized as potential bone imaging biomarkers in future CBCT radiographic studies of TMJ OA. It is foreseeable that as CBCT technology continues to evolve, trabecular characterization will be utilized for screening, diagnosis and monitoring of TMJ OA.

5. CONCLUSIONS

• Strong correlations between trabecular thickness, trabecular separation, bone volume, bone volume/total volume, bone surface area, and degree of anisotropy measurements in CBCT and micro-CT images were observed. This indicates that these parameters may be suitable bone imaging biomarkers that can be utilized clinically.

• Moderate correlations were found between osteoclast number and 3 trabecular bone structure measurements – connectivity, connectivity density, and the Euler characteristic. However, these 3 bone structure parameters did not significantly correlate to osteoclast density. Thus, this conclusion should be interpreted with caution as there are several confounding factors, which could have led to this result.

Statement of Clinical Relevance.

The analysis of the internal trabecular architecture of the resected condylar heads may be a method that helps to diagnose the disease early without having to wait until surface changes of cortical bone are visible that can help in the diagnosis of TMJ OA.