Effect of a prototype 2-dimensional antiscatter grid on image quality obtained with a dental cone-beam computed tomography scanner

Abstract

Purpose

X-ray scattering adversely affects cone-beam computed tomography (CBCT) image quality, generating image artifacts, causing inaccurate tissue density representation, and reducing contrast. This study evaluated the performance of a 2-dimensional antiscatter grid (2D grid) prototype in a dental CBCT system.

Materials and Methods

A focused 2D grid prototype was fabricated from tungsten and integrated with the detector of a dental CBCT system. Residual scatter transmitted through the 2D grid was corrected using a measurement-based scatter correction method. Phantom imaging experiments were performed in anatomical regions relevant to dental and head imaging with and without this grid. Following image reconstruction via filtered back projection, attenuation coefficients were converted to Hounsfield units (HU). Subsequently, scatter suppression performance, HU consistency, image artifacts, and contrast resolution were evaluated.

Results

The 2D grid reduced scatter intensity by a factor of 10–20 in CBCT projections. Consequently, the grid substantially increased contrast, reduced image artifacts, and improved HU consistency. The contrast increased by 27% and 48% in bone- and soft tissue-equivalent regions, respectively. HU value deviations among teeth decreased from 510 to 146 HU. These results indicate improved visualization and tissue density representation fidelity in CBCT images acquired with the 2D grid.

Conclusion

Use of a 2D grid could substantially improve the accuracy of tissue density representation and the contrast of dental and head anatomy in 3-dimensional images obtained with dental CBCT. Such improvements may translate to better quantitative evaluation of bone quality, enhanced tissue visualization, and more accurate model generation for surgical planning and guidance.

Introduction

Cone-beam computed tomography (CBCT) is a widely used imaging method in dentistry. In a CBCT scan, an X-ray source and a detector rotate around the patient to collect 2-dimensional (2D) X-ray projections, which are then reconstructed into high-resolution volumetric images. These images provide high diagnostic accuracy and assist in treatment planning. Despite the promise of dental CBCT, increasing image quality and quantitative accuracy remains a primary research area. In addition to qualitative improvements in tissue visualization and diagnostic performance, higher CT number accuracy is desired to extract quantitative information from dental CBCT images. Increased quantitative accuracy may expand the utility of dental CBCT in various applications, such as extracting bone density for dental implant placement, planning surgical tooth extractions, evaluating dental bone grafts, assessing trauma, and planning reconstructive or corrective jaw surgery. Another application that would benefit from improved quantitative accuracy is the generation of the 3-dimensional (3D) models used for surgical planning and fabrication of surgical guides based on CBCT images, which rely on accurate representation of tissue-specific CT numbers.1,2,3,4,5

To improve CBCT image quality, numerous technical challenges must be addressed, including the presence of artifacts from metallic objects, motion, beam hardening, image noise, and scatter.3,6 Among these issues, scattered radiation represents one of the most fundamental causes of image quality degradation in CBCT.7,8 Compared to a conventional multidetector CT system with a fan-shaped X-ray beam, a CBCT system employs a larger cone of X-rays in the cranio-caudal direction, increasing the fluence of scattered X-rays. In turn, scatter degrades contrast, increases image artifacts, and reduces CT number accuracy, which compromises the representation of tissue densities in CBCT images. Numerous methods have been investigated to address the scatter problem. The most common approach is algorithmic scatter correction, in which the scatter signal bias is corrected after the X-ray detector has captured scatter. Most scatter correction methods employ a model to estimate and correct the scatter bias from the CBCT raw data and subsequently compensate for the scatter signal.9,10,11 However, suboptimal modeling of scatter conditions may lead to misestimation of scatter intensity and incomplete scatter correction. Another approach for scatter suppression is the use of scatter rejection devices such as antiscatter grids, which prevent scattered X-rays from reaching the detector. Conventional radiographic antiscatter grids employ a 1-dimensional array of radiopaque septa that absorb incident scattered X-rays. Although conventional antiscatter grids reduce scatter intensity, their performance is insufficient to achieve highly accurate CT numbers in CBCT images.12,13 In addition to physics-based scatter correction and rejection techniques, image restoration methods have also been investigated, in which the effects of scatter are removed from CBCT images via heuristic and machine learning approaches.12,14,15,16

In the present work, a novel scatter suppression approach was investigated for dental CBCT systems. A prototype 2D antiscatter grid (hereafter termed the “2D grid”) was developed, composed of a 2D array of tungsten septa placed on the X-ray detector, with each septum aligned toward the X-ray source. Compared to conventional radiographic antiscatter grids with a 1-dimensional (1D) array of radiopaque septa, 2D grids can provide significantly improved scatter rejection performance and superior primary X-ray transmission. 17,18 While 2D grids have been shown to improve CBCT image quality in human torsos, they have not been investigated in the context of dental CBCT imaging.19,20,21 Due to the smaller X-ray field of view (FOV), shorter source-to-detector distance, and smaller anatomical regions imaged with dental CBCT systems, the scatter characteristics and resulting image quality improvements may differ from those in CBCT systems designed for imaging the torso. Hence, this study investigated the effect of 2D grid-based scatter suppression on dental CBCT image quality.

Materials and Methods

2D grid prototype

The prototype 2D grid (M2 Technologies, Denver, CO, USA) had a grid pitch of 1.8 mm, a grid ratio of 12, and a wall thickness of 0.1 mm (Fig. 1A). It was 3 cm wide in the axial direction and 25 cm long in the transverse direction of the CBCT system. Its grid walls were aligned towards the X-ray source in the source-detector geometry of the Planmeca Viso G5 CBCT system (Planmeca Oy, Helsinki, Finland). The grid was directly mounted on the X-ray detector (Fig. 1B). The prototype was manufactured from pure tungsten using powder bed laser melting process, with air in the spaces between the tungsten. Tungsten was chosen for its structural rigidity and superior X-ray attenuation properties compared to lead. Moreover, tungsten is well-suited for the powder bed laser melting process, which enables the fabrication of 2D grid structures with 0.1-mm-thick walls. Although lead is commonly used in conventional antiscatter grids, it is suboptimal for fabricating 2D grid structures using laser sintering-based additive manufacturing processes.

Fig. 1. A. Photo of the tungsten 2-dimensional (2D) antiscatter grid used in the experiments. B. 2D grid prototype as mounted on the X-ray detector.

While the 2D grid rejects the vast majority of scattered X-rays, a small fraction still passes through and adversely affects image quality. This residual scatter was corrected using a measurement-based scatter correction method, termed Grid-based Scatter Sampling (GSS).20,22,23 The GSS method utilizes the 2D grid itself as a scatter measurement device to directly measure and correct the remaining scatter in raw projections. Prior reports have demonstrated that the GSS method can accurately measure and correct residual scatter transmitted through the 2D grid.20,22,23 In this work, the GSS method was optimized and adapted for the 2D grid prototype and the dental CBCT system under investigation.

CBCT imaging experiments

The CBCT experiments were performed using a Planmeca Viso G5 dental CBCT system. While the 2D grid covered only a small section of the detector (Fig. 1B), the X-ray FOV covered the full active area of the detector to mimic scatter conditions in a dental CBCT scan. CBCT system and scan parameters are summarized in Table 1. The jaw protocol was used to acquire the data. The frames were acquired in an offset geometry setting (Fig. 2A). With this setting, the detector and the tube are offset toward the same side and rotate 360° around the object, such that one side at a time is visible to the detector. This allows a larger object to be reconstructed than under a symmetric geometry setting.

Table 1. Summary of imaging acquisition parameters.

Fig. 2. A. Diagram of the imaging system with the 2-dimensional (2D) grid and 3 sample projections of the anthropomorphic dental phantom used for scatter-to-primary ratio (SPR) calculations. B. SPR in CBCT projections acquired with the 2D grid in place. The average SPR of each pixel column in one of the projections is shown. C. Average SPR of the full phantom region. D. Average SPR in the low signal, or highly attenuating, region of each projection.

The effect of the 2D grid on CBCT image quality was evaluated in numerous phantom experiments, with the phantoms summarized in Table 2. For each phantom, 2 sets of experiments were performed: with and without the 2D grid. The no-grid and 2D-grid CBCT scans were acquired using identical dose and scan parameters, as summarized in Table 1. Projections acquired with the 2D grid were corrected for residual scatter using the GSS method. The system's default software-based scatter correction was applied to all datasets without the 2D grid. The correction algorithm attempts to remove smooth, low-frequency signal, also called cupping, from the soft tissue part of an image volume after reconstruction. Water-equivalent beam hardening correction was applied to both the 2D-grid and no-grid CBCT scans, and the images were subsequently reconstructed using the CBCT system software. Filtered back projection was applied to reconstruct all images.

Table 2. Summary of phantoms used in imaging experiments.

Effect of the 2D grid on scatter suppression and CBCT image quality

Scatter intensity reduction performance was evaluated in CBCT projections of an anthropomorphic dental phantom. The reduction in scatter intensity was characterized by the scatter-to-primary ratio (SPR) in projections acquired with and without the 2D grid. Raw CBCT projections of a phantom, Proj_phan, contain both primary and scattered X-rays. When CBCT projections are acquired with a 2D grid, a primary-only projection, Proj_GSS, was computed by correcting for scatter using the GSS method. Hence, scatter intensity (S) and SPR were calculated as follows:

S=Proj_phan-Proj_GSS (1)

$$\text{SPR}=\frac{{\text{Proj}}_{\text{phan}}-{\text{Proj}}_{\text{GSS}}}{{\text{Proj}}_{\text{GSS}}}$$ (2)

This approach was also used to estimate SPR in CBCT projections acquired without a 2D grid, since the primary signal intensity was the same in the 2D-grid and no-grid CBCT scans.

CBCT image quality was evaluated both qualitatively and quantitatively. The qualitative assessments focused on the image artifact reduction performance of the 2D grid, and they complemented and confirmed the quantitative analyses. For the quantitative evaluations, CBCT scans of each phantom acquired with and without the 2D grid were first rigidly co-registered; subsequently, Hounsfield unit (HU) consistency and contrast were analyzed. Image contrast was calculated by determining the difference between the target and background ROIs.

The image reconstruction process yielded a 3D map of linear attenuation coefficients. The attenuation coefficient for each voxel was converted to an HU value using the standard HU calculation method:24,25

$$\text{HU}\left(i,j,k\right)=\frac{\left(\mu \left(i,j,k\right)-{\mu }_{\text{water}}\right)}{{\mu }_{\text{water}}}\times 1000$$ (3)

where µ is the attenuation coefficient of a voxel located at i, j, k in the 3D volume and µ_water is the attenuation coefficient of water obtained from a calibration phantom. With this linear transformation, air corresponds to -1000 HU and water to 0 HU. The same HU calculation method was used for all imaging experiments.

First, HU uniformity performance was evaluated in CBCT images of a dental-focused anthropomorphic dental phantom (Erler-Zimmer GmbH, Lauf, Germany). Fourteen teeth were segmented from the co-registered images with and without the 2D grid using the MATLAB Image Segmenter (MathWorks, Natick, MA, USA). The ROIs for image analysis were the entire segmented teeth. An HU histogram of each segmented tooth was calculated, and the mean HU was analyzed to assess tooth-to-tooth HU variation.

Bony anatomy contrast was evaluated using the Sedentex CT IQ phantom (Leeds Test Objects, North Yorkshire, UK) with 4 calcium-containing cylinder modules from the Gammex 472 phantom (Sun Nuclear, Melbourne, FL, USA) inserted in an acrylic cylinder 160 mm in diameter. The signal difference between the high-contrast inserts and the acrylic background was computed to compare the bony anatomy contrast in 2D-grid and no-grid CBCT images. The ROIs for the calcium cylinders were circular, with a diameter equal to half the diameter of the cylinders (approximately 17 mm). A similarly sized circular ROI was used for the acrylic background near the calcium contrast inserts.

Due to the lack of soft tissue contrast in the other dental phantoms available at the time of this study, soft tissue contrast performance was evaluated using the brain portion of the CT Torso Phantom CTU41 (Kyoto Kagaku, Kyoto, Japan). For simplicity, this phantom will be referred to as the “brain phantom” for the remainder of this paper. ROIs were selected to calculate signal differences for the gray matter/ventricle and gray matter/hemorrhage regions. HU uniformity was also evaluated in the brain phantom. The skull and brain regions of the phantom were segmented by thresholding. HU uniformity in soft tissues was evaluated by placing 9 ROIs across the brain soft tissue region, while HU uniformity in bony tissues was evaluated by placing 16 ROIs across the brain skull region. The standard deviation and maximum deviation with respect to the mean HU across all ROIs were investigated. The shape and location of the ROIs are shown in the figures in the Results section.

Results

Scatter-to-primary ratio

Example CBCT projections of the anthropomorphic dental phantom acquired with the 2D grid, along with the 2D grid imaging diagrams, are presented in Fig. 2A. Without the 2D grid, SPR values varied between 0.5 and 2, with the highest values observed in the central section of the phantom where the X-ray beam exhibited the greatest attenuation. With the 2D grid in place, SPR values fell below 0.1 across the entire phantom projection (Fig. 2B). On average, the 2D grid reduced SPR by a factor of 10 in each projection (Fig. 2C). In highly attenuating regions of the projections, SPR was reduced by up to a factor of 20 on average (Fig. 2D).

HU uniformity and image contrast improvements

Images of the dental phantom acquired with and without the 2D grid are provided in Figures 3A and 3B. While both images appear qualitatively similar, the CT numbers for dentin were substantially higher in images acquired with the 2D grid due to X-ray scatter suppression. The mean HU values of all teeth were 1718 HU with the 2D grid and 1227 HU without it. HU values also varied as a function of tooth location, as seen in segmented images (Figs. 3C and 3D). HU histograms of teeth #2 (maxillary right second molar), #6 (maxillary right cuspid), and #9 (maxillary right central incisor) under the Universal Numbering System are shown in Figure 4. In images acquired without the 2D grid, the HU histograms exhibited large differences among teeth. With the 2D grid, these differences were significantly reduced due to robust scatter suppression. Specifically, the standard deviation of HU values among teeth decreased from 166 to 43 HU, the maximum deviation from the mean decreased from 275 to 116 HU, and the maximum difference among teeth decreased from 510 to 146 HU.

Fig. 3. A. Axial image with the 2-dimensional (2D) grid (window width/level=4000/1000). B. Axial image without the grid (window width/level=4000/1000). C. Magnified section of teeth with the 2D grid (window width/level=1000/1800). D. Magnified section of teeth without the grid (window width/level=1000/1300).

Fig. 4. Sample Hounsfield unit (HU) histograms of the segmented teeth in images acquired with (A) and without (B) the 2-dimensional (2D) grid. The histograms represent the maxillary right second molar (#2), maxillary right cuspid (#6), and maxillary right central incisor (#9). C. Mean HU value of each segmented tooth based on the Universal Numbering System.

Bony anatomy contrast was evaluated using calcium-containing inserts placed in the Sedentex phantom. Contrast increased linearly as a function of calcium concentration for both 2D-grid and no-grid configurations (Fig. 5). For a 600 mg/mL calcium insert, the measured contrast was 1782±143 HU in images acquired with the 2D grid and 1345±117 HU in images acquired without it.

Fig. 5. Hounsfield unit (HU) values of calcium-containing inserts. 2D: 2-dimensional.

Figure 6 shows images of the brain phantom captured with the dental CBCT system. The contrast between structures such as ventricles and hemorrhage, relative to the background brain parenchyma, appeared noticeably greater in images acquired with the 2D grid. CT number line profiles in Figure 7 also demonstrate superior soft tissue contrast with the 2D grid. Quantitatively, soft tissue contrast increased from 19–62 HU to 27–89 HU, corresponding to an average increase of 48% (Table 3).

Fig. 6. Brain phantom image slices acquired with (A) and without (B) the 2-dimensional (2D) grid. Image slices depict the hemorrhage acquired with (C) and without (D) the 2D grid. Display window width/level: 100/60.

Fig. 7. Line profiles drawn over ventricle (A) and hemorrhage (B) according to the regions of interests (ROIs) indicated in C and D, demonstrate soft tissue contrast in images acquired with and without the 2-dimensional (2D) grid.

Table 3. Soft tissue contrast across ventricles and hemorrhage in the anthropomorphic brain phantom (unit: Hounsfield unit).

ROI: region of interest, 2D: 2-dimensional

Figure 8 illustrates the effect of the 2D grid on shading artifacts. In the region indicated by the red box in Figure 8A, significant shading artifacts were observed near the skull in images acquired without the 2D grid (Fig. 8B), and these artifacts were markedly reduced in images captured with the grid (Fig. 8C). Similar effects are also evident in another image slice (Figs. 8D-F). These observations were further supported by the HU uniformity among ROIs placed in the brain parenchyma (Fig. 9). When the 2D grid was applied, the standard deviation of ROI mean HU values decreased from 4.8 to 2.4 HU. HU uniformity in the ROIs positioned in the skull region also improved upon application of the 2D grid (Fig. 10), with the standard deviation of ROI mean HU values decreasing from 208 to 104 HU.

Fig. 8. Magnified sections of the artifact region indicated by the white box in A are shown in B acquired with the 2D: 2-dimensional (2D) grid and C (without the 2D grid). Similarly, magnified sections of the artifact region indicated by the white box in D are shown in E (acquired with the 2D grid) and F (without the 2D grid). 2D: 2-dimensional.

Fig. 9. Mean Hounsfield unit (HU) values in 9 region of interests (ROIs) demonstrate improved HU uniformity in the brain phantom. The ROI locations are shown in the inset. With the 2-dimensional (2D) grid, the standard deviation of HU values among the ROIs was reduced from 4.8 to 2.4 HU. The maximum deviation from the mean HU of all ROIs was 8.3 HU without the 2D grid and 4.5 HU with the grid.

Fig. 10. Mean Hounsfield unit (HU) values in 16 region of interests (ROIs) demonstrate improved HU uniformity in the skull region. The ROI locations are shown in the inset. With the 2-dimensional (2D) grid, the standard deviation of HU values among the ROIs was reduced from 208 to 104 HU. The maximum deviation from the mean HU of all ROIs was 144 HU without the 2D grid and 64 HU with the grid.

Discussion

This study investigated the effect of robust scatter suppression on CBCT image quality in a dental CBCT system. The prototype 2D grid demonstrated high efficacy in rejecting scattered X-rays in dental CBCT geometry, achieving a reduction in SPR by a factor of 10 to 20 in large-FOV CBCT scans.

In all phantom experiments, the use of the 2D grid resulted in consistent and uniform HU values for a given material type, and contrast was substantially improved, particularly in bony anatomy. These findings indicate that image artifacts, contrast loss, and HU degradation caused by scatter are significantly reduced by the 2D grid under the same X-ray exposure level as scans captured without the grid. This effect occurs because in projections, bony tissue regions often exhibit a higher SPR than soft tissue regions. Bony anatomy strongly attenuates primary X-rays due to its higher atomic number and density, while the scatter signal remains relatively uniform. Consequently, the rapid reduction of primary signal in bony regions yields a higher SPR and greater CT number degradation.10,26 When scatter is effectively suppressed, improvements in CT number consistency are more pronounced. Although the scatter transmission properties of the 2D grid do not depend on the exposure level, the rejection of scatter combined with exposure level variations may impact image noise characteristics. The removal of scatter improves contrast in CBCT images, but it also increases noise due to the reduced overall X-ray fluence reaching the detector. The interplay between increased noise resulting from reduced scatter and exposure level variation remains an area for further investigation.

Another area of interest is the accuracy of HU values in high-density regions, such as dentin, after scatter suppression. However, estimating the ground truth HU values for dentin is nontrivial because HU values vary strongly as a function of X-ray energy. The mean energy of the 110 kVp X-ray spectrum used in this study was estimated at 65 keV in the absence of a phantom; however, beam hardening in dentin would likely increase the mean beam energy to between 70 and 90 keV, corresponding to a theoretical dentin HU range of 1520–2240 HU. The mean HU value of dentin was 1718 HU in CBCT images acquired with the 2D grid, which falls within the expected range, whereas it was 1227 HU in images without the 2D grid, indicating an underestimation of dentin density. Contrast was also improved in both soft and hard tissue regions. For example, contrast in calcium-containing regions was 27% higher (Fig. 5), and soft tissue contrast in a brain-mimicking phantom was approximately 50% higher (Fig. 7).

Improved CT number accuracy in bony anatomy and teeth has various quantitative imaging applications, including facilitating the extraction of bone density from CBCT images for dental implant placement planning, surgical tooth extractions, evaluation of dental bone grafts, and corrective jaw surgery. Virtual 3D models generated for surgical planning rely on precise segmentation of anatomical structures; improved CT number accuracy may enhance segmentation and thus yield more accurate 3D models. In implant treatment planning, better CT number accuracy can assist in quantitatively assessing bone quality at an implant site. Regarding post-implant CBCT images, more accurate representation of bone density and morphology may improve evaluations of peri-implant bone, such as assessments of bone integration or bone loss.5 Additionally, higher soft tissue contrast could facilitate the visualization of soft tissue lesions and gingival soft tissue thickness.1 Although the dental CBCT system used in this work is not designed for intracranial imaging, the observed improvement in soft tissue contrast suggests that a flat-panel detector-based CBCT system integrated with a 2D grid could potentially be used beyond dental and maxillofacial imaging, such as for bedside imaging of hemorrhage in the brain.27

Another area of interest is the comparison of CBCT images acquired with the 2D grid prototype to those obtained using a conventional 1D antiscatter grid, which consists of a 1D array of lead lamellae. Although 1D grids are commonly used in diagnostic X-ray imaging,28 a conventional 1D grid with a focusing geometry suitable for dental CBCT systems was not available for this study. Previous studies comparing 2D and 1D grids in CBCT systems—primarily designed for imaging the human torso. These works demonstrated that 2D grids have a factor of 2–6 lower scatter transmission than conventional 1D grids.17,18 In general, 2D grids provide superior scatter suppression, CT number consistency, and contrast-to-noise ratio than 1D grids when imaging large FOVs or anatomical regions, such as the pelvis.20 An image quality comparison of 1D and 2D grids in dental CBCT remains a subject for future investigation.

This study had several limitations. First, the 2D grid prototype was relatively small, covering only about 18% of the FOV in the cranio-caudal direction. A larger 2D grid spanning the full FOV of the CBCT system remains to be investigated. Although the grid prototype in this initial feasibility study covered only a small section of the X-ray detector, the full FOV was irradiated in all experiments, as in a full-FOV CBCT scan. This ensured that the scatter intensity generated by the imaged object was equivalent to that in a full-FOV scan, thus mimicking realistic scanning conditions. It is expected that a larger 2D grid would exhibit similar scatter transmission characteristics and a comparable trend in image quality improvement. The impact of scatter mitigation on image quality using smaller FOV protocols will be investigated in the future. Second, the image processing software was not fully optimized for 2D grid implementation. The noise properties of CBCT images acquired with the 2D grid differ substantially due to its X-ray transmission characteristics and require further research. Third, this study did not investigate the effect of robust scatter suppression on metal artifacts.

In this work, a prototype 2D grid was developed and evaluated for dental CBCT systems. The prototype reduced scattered X-ray intensity and improved the fidelity of CBCT images. Notably, a significant improvement in HU value uniformity was observed in bony structures and teeth. Such improvement could enhance the utility of dental CBCT imaging for the quantitative evaluation of bone density, provide better visualization of peri-implant tissues, and enable more accurate generation of 3D models for surgical planning and surgical guides. Improved contrast and reduced artifacts in soft tissues may further expand the applications of dental CBCT systems in soft tissue imaging. Future work will focus on optimizing 2D grid properties for dental CBCT and fabricating a larger grid prototype to enable imaging of larger anatomical regions.