Bone is a connective tissue designed to bear loading generated by muscle movement and body weight during daily activity.[1,2] Bone in a live skeletal system constantly changes. Approximately 7.5% of bone turnover occurs annually.[3] Bone modeling and remodeling are inherently involved in the alteration of bone. In the process of bone remodeling, a bone resorbing cell, osteoclast, is activated to remove the pre-existing bone tissue and subsequently, new bone is deposited by a bone forming cell, osteoblast.[3,4,5,6,7] The bone modeling is an uncoupled process consisting of activation and resorption or activation and formation. When bone resorption and formation are balanced, the net quantity of bone is maintained. It has been observed that bone quantity is associated with its mechanical strength.[8,9] As bone in a live patient cannot be subject to mechanical testing, many non-invasive imaging technologies have been developed to estimate the bone quantity as a surrogate for a patient's bone fragility. Computed tomography (CT) has been widely used non-invasive device to assess bone mass.[10,11] This X-ray based tool mainly detects mineral, which is the densest component in bone tissue materials, and provides Hounsfield units (HU) as the measures. It is well known that bone is composed of approximately 30% water, 40% mineral, and 30% organic protein (mostly collagen) at the tissue level.[1] In the initial process of bone formation, osteoblasts produce immature collagen and mineral composite.[12] The mineral content rapidly increases up to 70% of full mineralization within one week post-initiation of bone formation.[13] Following the first mineralization, more minerals are added to the collagen-mineral composite during the secondary mineralization process that lasts for years.[13,14] Bone modeling and remodeling processes are activated at different time points. In particular, active bone remodeling is stimulated by bone disease and complications including estrogen deficiency in postmenopausal osteoporosis, osteoarthritis, fracture healing, and implantation surgery.[2,6,7,15,16,17,18] As a result, heterogeneity of bone tissue mineral density (TMD) increases (Fig. 1).

BMD accounts for bone quantity as the mineral contents within an apparent volume of bone, which includes porosity, bone marrow, as well as bone matrix. On the other hand, the TMD represents mineral contents only in the matrix of bone.[18,19] Bone quantity measurements have been widely used to diagnose bone disease. For example, according to the World Health Organization (WHO) criteria, osteoporosis is defined as 2.5 standard deviations or more below the average value for young healthy women.[20,21] However, it has been indicated that bone quantity alone cannot explain bone fragility.[20,22] On the other hand, the bone quality is assessed by its morphology and TMD distribution.[23,24] Mechanical properties of trabecular bone are explained using three-dimensional (3D) morphological parameters of trabecular bone such as trabecular number, thickness, and anisotropy ratio.[25,26,27] Recently, many studies have also indicated that TMD is responsible for controlling elastic, plastic, and viscoelastic mechanical properties at the tissue level of bone.[28,29,30,31] As such, efforts have been made to develop a new imaging technology that can non-invasively assess the bone quality. Dual energy X-ray absorptiometry (DXA) has been recognized as a standard technique to obtain the BMD of a patient's spine or hip clinically. However, the DXA produces a low resolution (approximately 500 µm [32]) 2D image that only provides areal BMD and cannot precisely delineate cancellous bone structure with the trabecular thickness level of 50 to 200 µm.[25,33,34] A high-resolution (up to less than 1 µm) 3D micro-CT image can describe detailed bone morphology and TMD more accurately than other CT images (Fig. 2). However, much higher radiation doses are generated during the longer scanning time of micro-CT for the same size of a specimen, which limits its use to laboratory research purposes only. On the other hand, cone beam CT (CBCT) provides higher image resolution with less radiation dose and exposure time than the conventional clinical multidetector CT (MDCT) (Table 1). Thus, CBCT has been widely used for the clinical diagnosis of dental complications.[35,36] The 3D image produced by CBCT allows for detailed morphological analysis of bone. However, reliability of CBCT based bone density measurement remains to be determined.[37,38,39] Thus, this review will discuss whether the CBCT can be used to assess BMD at next sections.

Noise, ring and streak artefacts can be corrected by improving reconstruction algorithms.[39,42,49] Recently, a scatter-correction algorithm was developed to significantly improve the consistency of CBCT gray values including compressed sensing reconstruction algorithms to create a scatter-free CBCT image.[50,51] As the beam hardening effect is the most frequently observed artefact in CT scanning, many efficient correction algorithms have been developed and included in commercial CT software.[52,53] The FOV associated variations may arise because the backprojection algorithm includes gray values of tissues outside the small FOV during reconstruction.[39] Taken together, the recently improved reconstruction process likely corrects the local errors of gray value in the CBCT image. However, as these advanced algorithms need to be verified and calibrated to individual CT systems prior to being utilized in the clinical settings, the artefacts may be still observed in most of the currently used CT machines.

While the local errors can be visually detected, there are some systematic complications to be considered for CBCT based mineral density measurement. The most debated aspect is that the HU values of subjects are not consistent between different CT systems and between different times scanned even using the same CT system. These discrepancies can arise from the non-uniform process of scaling the HU values during reconstruction. It was indicated that a manufacturer's software imports a gray value range of 1 to 4,096 and rescales from -1,024 to 3,072.[47] This image would have 12-bit data size (2¹²=4,096). The gray value of an image with 8-bit data size is scaled from 0 to 255 (2⁸=256). The HU of water and air should be close to 0 and -1,000, respectively. However, the different scaling processes may provide completely different absolute values of HU for the same material. As such, the value of each voxel in a CBCT image may better represent the gray value rather than the HU.

Many studies found that strong positive correlations of the gray values of CBCT image with known density of reference materials and gray values obtained from the conventional clinical CT.[47,48,54,55,56,57,58,59] For example, phantoms of bone materials (hydroxyapatite) with 3 different densities (1,000, 1,250, and 1,750 mg/cm³) were scanned using 3 different resolutions (200, 300, and 400 µm) of CBCT (Fig. 4). As a result, the CBCT based gray values of phantoms had strong positive correlations with the density values. The gray value of scanned specimens can be directly converted to the corresponding mineral density value using this calibration curve when the same CBCT scanner is used. Alternatively, a previous study found that the CBCT gray value strongly correlated with the CT attenuation coefficient (µ) of standard materials, which include aluminium (µ=0.678), outer bone equivalent (µ=0.523), inner bone equivalent (µ=0.255), polymethylmethacrylate (PMMA) (µ=0.220), muscle equivalent (µ=0.205), water equivalent (µ=0.202), adipose equivalent (µ=0.176), and air (µ=0.000).[47] Then, the corrected HU value of scanned materials was derived by estimating the attenuation values in the HU equation (Eq. 2) using the correlation equation with the CBCT gray values. The corrected HU values were comparable between different CBCT scanners. Recently, clinical applicability of this HU correction was validated using intraoral phantoms for patients.[48]

The aforementioned methods need external standard phantoms to be scanned to obtain the calibration curve. Another methodology was introduced using an internal reference to compare the mineral density between patients.[60,61] In the previous study using CBCT images of human cadaveric mandibles,[61] teeth were digitally separated from the mandible segment using imaging software (Livewire®,[62] Institute of Computing, State University of Campinas, Brazil[61] and ImageJ, National Institutes of Health, Bethesda, Maryland, USA) (Fig. 5). The alveolar bone (AB) region was digitally isolated from within 1 mm of the root surface and the basal control bone (CB) was determined at 0.6 mm from periosteal and endosteal surfaces of the basal bone. The gray value histogram was used to obtain mean, standard deviation (SD), coefficient of variation (COV=SD/Mean), low and high gray values (Low5 and High5) at the 5th and the 95th percentiles of voxel counts in the histogram. The gray value parameters of the CB region were used as the internal references to compare those of AB region between subjects. It was assumed that a core region of basal cortical bone has normal bone remodeling while the AB and marginal basal bone regions have a rapid bone turnover due to active bone formation.[17,63] Percentage difference (%) of these gray level parameters between AB and basal cortical bone ([AB-CB]/[(AB+CB)/2]×100) was then computed for each patient CBCT image. The same image process was performed using the micro-CT images of the same human mandibles. As a result, the relative values of percentage differences were comparable between CBCT and micro-CT images based analyses. Those procedures to assess the gray value parameters were also applied to examine CBCT images collected from routine patients in clinic. An advantage of this method is that the relative value of the percentage difference is computed using the gray values of different regions in the same CBCT image avoiding the effects of different scanning conditions. The gray value histogram analyses were successfully used for clinical investigation of the success and failure of dental grafting and implant systems.[64,65]

To date, CBCT based BMD measurement has been used mainly to estimate bone properties for dental implantation.[66,67,68,69,70] The BMD measured from CBCT images showed a better predicting power to estimate cortical bone fracture than assessed by the conventional DXA images.[71] More applications are performed for the diagnosis of oral bone disease, improvement of orthodontic treatment, and maxillofacial surgery.[36,72,73,74] It is clear that higher resolution of 3D images can provide more useful information. For example, the partial volume effect can be reduced using a smaller voxel size in the image. However, if longer scanning time is required to obtain higher image resolution, the risk of high radiation exposure likely increases against the principle of "as low as reasonably achievable" (ALARA) radiation dose. As complete segmentation of bone voxels from non-bone voxels in the CT image with the current resolution can tremendously improve the capability of CT based analyses for both morphology and density (Fig. 2), numerous studies have been attempted to develop innovative algorithms.[75,76,77,78] Most commercialized CBCT software uses a global thresholding gray value to segment voxels. However, this simple method does not provide sufficient quality of images.[79] Further studies are still required to develop an optimal segmentation technology specific to individual CT images.

Mineral density distribution of bone tissue reflects the result of biological activity, which is altered due to bone complications. Clinical CT is an indirect method to measure mineral density distribution based on the X-ray attenuation coefficient of the materials, mainly the mineral, in bone tissue. Hence, CT based density measurement is not as accurate as direct measurement using biopsy. However, it is a very powerful non-destructive tool that allows for longitudinal diagnoses of patients' bone disease. Most of the local artefacts can be corrected by improving the CBCT image reconstruction algorithm and alternative corrective methods can be applied as visualized in the image. However, the image processing errors during converting the attenuation coefficient values to the HUs and gray values should be accurately addressed in order to compare the mineral density in CBCT images scanned under different conditions. This review briefly introduces the types of artefacts that occur and solutions to correct them as a quick reference for users who may find these errors in CBCT images.