A dual-energy CBCT with reduced scatter and cone beam artifacts using an X-ray source array and interlaced spectral filters

Abstract

Objective:

To design a dual-energy cone beam computed tomography (DE-CBCT) scanner with reduced scatter and cone beam artifacts.

Methods:

The scanner designed for maxillofacial imaging comprises a carbon nanotube (CNT) X-ray source array with multiple focal spots (“sources”) and an energy integrating flat panel detector (FPD). The X-ray photons from each focal spot were narrowly collimated in the axial direction and was filtered by interlaced low- and high-energy spectral filters. Two sets of projection images were acquired by sequentially activating the X-ray beams from each source in one gantry rotation. The projections were processed using a one-step inversion algorithm. An anthropomorphic head phantom, a Defrise phantom and a water-equivalent phantom containing calcium and iodine inserts were used to compare the performance of the new dual-energy multisource CBCT (DE-MS-CBCT) with a conventional DE-CBCT using the same air-kerma.

Results:

The DE-MS-CBCT eliminated the cone beam artifacts, reduced the degree of cupping artifacts from 14.53% to 2.94%, and lowered the mean relative error of water density from 15.3% to 1.7%, while the accuracies for iodine and calcium densities were comparable. The contrast-noise-ratios (CNR) of the calcium and iodine inserts against the solid water increased by 4.8%–53.4%.

Conclusion:

The DE-MS-CBCT reduces scatter and cone-beam artifacts, increases the image CNR, and enhances accuracy of materials quantification without increasing X-ray exposure compared to the conventional DE-CBCT.

Significance:

The results demonstrate a new DE-CBCT method with improved image quality and accuracy of materials quantification without the need for an energy sensitive detector or kV switching.

I. Introduction

CONE beam computed tomography (CBCT) is used for a wide range of applications, including image-guided radiation therapy, interventional radiology, surgical planning, and extremity, breast and imaging [1–3]. It provides 3D images with high isotropic resolution at reduced patient radiation exposure, lower cost, and smaller footprint compared to multi-detector CT (MDCT). A CBCT scanner typically comprises an energy-integrating flat-panel detector (FPD) and an X-ray source that emits a polychromatic beam with a wide cone angle. It captures the full volumetric data in one gantry rotation without the need for patient translation. By utilizing two different X-ray spectra, dual-energy CBCT (DE-CBCT) can extracts additional information and improves the performance compared to single-energy CBCT[4–6]. It enables material quantification and the generation of virtual monoenergetic images (VMI).

Various strategies have been investigated for DE-CBCT imaging, including the use of two source-detector pairs[7], multiple X-ray tubes and a single detector[8], kV switching[9, 10], a dual-layer detector[11–13], and one source-detector pair in two scans at different energies. Spectral filtration of the X-ray source is a relatively economical way to generate two distinct energy spectra. It has been implemented in a commercial DECT[14, 15] and investigated for DE-CBCT[9, 16–18]. We recently demonstrated DE-CBCT imaging by spectral filtration of an X-ray source with two focal spots[19, 20]. Because of the large imaging volume, strong scatter and cone beam artifacts ubiquitous in CBCT remain in the current DE-CBCT. Scatter leads to large errors and spatial nonuniformity (cupping artifact) of the HU values[21], degrades materials quantification, accuracy and reduces the image contrast resolution. Cone beam artifacts cause image distortion away from the central plane.

Various software-driven and hardware-based approaches have been investigated to reduce scatter and cone-beam artifacts in CBCT and DE-CBCT[1, 2, 11–13, 22–24]. CBCT architectures using multiple X-ray sources have also been reported. A discontinued commercial extremity CBCT used three axially positioned X-ray tubes, each covering the entire field of view (FOV), which was reported to reduce the cone beam artifacts[25]. Replacing the single wide cone angle X-ray tube in the conventional CBCT with multiple narrowly collimated X-ray beams in the axial direction was found to reduce scatter and cone beam artifacts in numerical and experimental simulations[26–28]. We recently demonstrated a multisource CBCT (ms-CBCT) using a carbon nanotube (CNT) based field emission X-ray source array with an external collimator [29, 30]. A stack of narrowly collimated and rapidly scanning X-ray beams was generated by sequentially activating the radiation from each focal spot. Phantom imaging studies show the design significantly reduces scatter, essentially eliminates cone beam artifacts, improves contrast-noise-ratio (CNR), and increases spatial uniformity and accuracy of the HU values compared to the current clinical dental CBCT[28, 29, 31].

The purpose of this study was to design a DE-CBCT scanner using multiple collimated X-ray beams and interlaced spectral filters without using either kV switching or an energy sensitive detector. The aim was to improve the image quality and accuracy of DE-CBCT without increasing the imaging dose compared to the current scanners. A one-step materials decomposition and reconstruction algorithm[32, 33] was modified for this particular imaging configuration. The imaging performance was evaluated using phantoms and compared to a DE-CBCT operated in the conventional configuration and similar dose.

II. Methods

A. Imaging System

An existing benchtop single energy ms-CBCT designed for maxillofacial imaging[29] was modified for this purpose. The system includes a linear CNT-based field emission X-ray source array positioned along the axial direction and a FPD, both mounted on a rotating gantry (Figure 1). The source-to-object distance of 400 mm and source-to-detector distance of 615 mm are similar to those of a commercial dental CBCT (CS9300, Carestream Dental). The detector (Xineos-1511, Teledyne Dalsa, Waterloo, CN) with an active area of 147.1 mm (width) × 113.7 mm (height) and a pixel size of 99 μm was shifted laterally by 70 mm to extend the field-of-view (FOV) to 187 mm (width) × 90 mm (height) at the rotation center. The detector offset configuration is used in several commercial dental CBCT for cost saving purpose [34]. The detector was operated in a 2×2 binning mode. The CNT X-ray source array[35, 36] (NuRay Technologies, Changzhou, China) comprises eight individually programmable CNT field emission cathodes, gate electrodes, and an extended tungsten anode enclosed in a stainless container with a 1.7 mm thick aluminum X-ray window, as described previously[30]. The X-ray focal spots are evenly spaced with a 12 mm inter-source spacing. The array was operated at 110 kV, in the range of what is commonly used for patient maxillofacial imaging by clinical density CBCT (90-120kV)[37]. The exposure sequence, time, and tube current controlled by an electronic control module (ECS, NuRay Technologies).

Figure 1.

(a) An illustration of the method for DE-MS-CBCT imaging by spectral filtration of a X-ray source array. (b) A photo of the benchtop scanner. (c) A photo of the CNT X-ray tube with the external collimator. (d) A cross-sectional illustration of the multi-source collimator configured for dual-energy imaging.

To generate dual-energy X-ray beams, the radiations from four even number focal spots were filtered by a low-energy (LE) filter of 0.15 mm thick Ta, and the radiations from four odd-number focal spots by a high-energy (HE) filter of 0.5 mm Sn, in addition to the inherent filtration from the 1.7 mm Al X-ray window. The filter materials and material thickness were optimized by spectrum simulation using the opensource software Spektr[38] and experiments, as previously reported[19, 20]. This produces a mean energy of 60.4 keV and half-value-layer (HVL) of 8.23 mm Al for the LE beams and a mean energy of 75.9 keV and HVL of 10.5 mm Al for the HE beams[20]. The simulated high- and low-energy spectra are demonstrated in Figure 2.

Figure 2.

Simulated spectra for the high-energy (HE) and low-energy (LE) beams. Each spectrum includes the 1.7 mm Al inherent tube filtration plus an additional 0.5 mm Sn filter for HE or 0.15 mm Ta filter for LE, yielding an energy separation of 15.5 keV.

A new multisource collimator with eight trapezoid apertures, shown in Figure 1(c, d), was designed to confine the radiation from each focal spot to cover only a section of the object in the axial direction. The four collimated beams from each energy group collectively illuminated the entire FOV, as illustrated in Figure 1(a). The adjacent beams in the same energy group were designed to intercept at 80 mm before the isocenter to ensure complete sinogram coverage for maxillofacial imaging. This leads to a cone angle of 4.14° for each beam, compared to 10.5° for the regular DE-CBCT geometry using a single X-ray source covering the entire detector under the same conditions. The fan angle of each beam is 13.53°, providing full coverage of the detector width. The collimator was manufactured out of stainless steel.

B. Phantoms

A 160 mm diameter cylindrical water-equivalent plastic phantom (SolidWater, Sun Nuclear Co, Melbourne FL) with multiple wells (28.5 mm in diameter and 80mm in depth) filled with either iodine inserts with the concentrations of 2, 5, 10, and 15 mg/ml (GAMMEX, Sun Nuclear Co, Melbourne, FL) or calcium inserts with the respective concentrations of 50, 100, and 300 mg/ml, Figure 1 (b), was imaged in separate scans. To evaluate cone beam artifacts, a homemade Defrise phantom, Figure 3 (a), consisting of a stack of acrylic discs (160 mm in diameter × 4 mm thick) with 4 mm air gap was imaged. An anthropomorphic adult skull and tissue-equivalent head phantom (RANDO, Nuclear Associates, Hicksville, NY) was also evaluated, Figure 3 (b).

Figure 3.

(a) A homemade Defrise phantom. (b) An anthropomorphic adult head phantom (RANDO, Nuclear Associates).

C. Imaging Protocol

For dual-energy multisource CBCT (DE-MS-CBCT) imaging, the eight individual sources were cycled sequentially 360 times during a single 360° gantry rotation, yielding 2880 projection frames (1440 per energy). The exposure settings are listed in Table 1. For comparison, the phantoms were also scanned using the same scanner in the conventional DE-CBCT geometry (referring to as “DE-CBCT”), with one source operated at LE and the other at HE. A different collimator was designed and applied such that each beam covers the entire FOV[20]. The datasets from both DE-CBCT and DE-MS-CBCT were reconstructed using a one-step inversion algorithm.

Table 1.

Exposure Parameter of DE-CBCT and DE-MS-CBCT

	DE-CBCT		DE-MS-CBCT
	LE	HE	LE	HE
Tube Voltage (kVp)	110	110	110	110
Filtration	0.15 mm Ta	0.5 mm Sn	0.15 mm Ta	0.5 mm Sn
Mean Energy (keV)	60.4	75.9	60.4	75.9
X-ray Cone Angle	10.5°	10.5°	4.14°	4.14°
Number of Beams	1	1	4	4
Tube Current (mA)	15	15	11	12
Exposure Time (ms)	5.0	5.0	5.0	5.0
Number of Views	360	360	360	360
Exposure (mAs)	27	27	19.8	21.6
Average Air-Kerma Rate (μGy/s)	1067	584	721	407
DAP (mGy·cm2)	321.2	175.8	394.0	221.6
Average Air-Kerma* (mGy)	2.45	1.34	2.34	1.31
Effective Field of View	187 mm (w)× 70 mm (h)		187 mm (w)× 90 mm (h)

^*Average air-kerma at the isocenter, calculated from DAP measured in air at the detector surface and divided by the effective FOV.

D. X-Ray Air-Kerma

A dosimeter (Raysafe X2, Unfors Raysafe AB, Sweden) was placed at the center of the radiation field of each beam on the detector surface, with no object present, to measure the X-ray air-kerma rate. The dose area product (DAP) of the scan at each energy was then calculated from the measured air-kerma rate by:

$$DAP=D_p\times L_w\times L_h\times N_{\mathit{\text{view}}}\times N_{\mathit{\text{source}}}$$ (1)

where, D_p is the air-kerma per X-ray pulse measured at the detector center. L_w and L_h are the width and height of the area directly illuminated by each beam on the detector. The measured value (full width at half maximum) for L_w is 147.1 mm, and L_h is 51.6 mm for the LE beam and 51.4 mm for the HE beam. N_view is the number of projections and N_source the number of X-ray sources at each energy. The total DAP of a dual energy scan is the sum of the DAP at the two energies. The average air-kerma at the rotation center was estimated by dividing the DAP by the area of the effective field of view.

E. Residual Scatter Subtraction

The residual scatter produced within the narrow imaging zone in each exposure was estimated and removed by an Adjacent Scatter Ratio Subtraction (ASRS) method which was developed and found to be effective for the ms-CBCT geometry[28, 29]. The scatter recorded on the detector rows not exposed by the primary beam was used to estimate and remove the scatter in the exposed area. The adjacent scatter ratios at different detector columns were calculated and fitted by a spline function to obtain the smoothed ratio, which was used to estimate the intensity of the primary photon.

F. Materials Decomposition and Reconstruction, and Virtual Monoenergetic Images

Projection-domain decomposition requires registration of the low and high energy images[39], and image-domain decomposition is not effective in mitigating beam hardening effects and has high noise[40]. Advanced one-step decomposition approaches have been developed that mitigate these limitations[32, 33, 41]. By integrating the forward model and acquisition geometry into a single framework, they can explicitly model non-coincident low- and high-energy ray paths[42], effectively reduce beam hardening, and can achieve a lower noise level compared with the conventional methods. In our previous study[20], a one-step inversion algorithm originally developed by Mechlem and coworkers for fan-beam geometry[32, 33] was modified for the cone-beam system with two axially offset X-ray beams using the open-source ASTRA toolbox[43, 44].

According to Beer-Lambert’s Law:

$$N^{LE,HE}=\int N_0^{LE,HE}{S}^{LE,HE}(E)exp(-A\mu (E))dE$$ (1)

where N^LE,HE, are the measured counts from two energy channels after attenuation, $N_0^{LE,HE}$ are the counts without attenuation, S^LE,HE (E) are the energy spectra, A is the system matrix and μ(E) is the linear attenuation coefficient. A linear combination of two base materials can be used to express μ(E):

$$\mu (E)={\left(\frac{\mu }{\rho }\right)}_1^E{\rho }_1+{\left(\frac{\mu }{\rho }\right)}_2^E{\rho }_2,$$ (2)

where, ρ_1,2 are the known densities of the base materials, and ${\left(\frac{\mu }{\rho }\right)}_{\mathrm{1,2}}^E$ are the mass attenuation coefficients that were obtained from the X-Ray Mass Attenuation Coefficient database maintained by the National Institute of Standards and Technology (NIST).

The proposed one-step inversion utilizes a negative log-likelihood function to measure how well the excepted data $\bar{N}$ agree with the measured data N:

$$-L\left(\bar{N}\right)=-Nlog\bar{N}+\bar{N},$$ (3)

To reduce noise and preserve edge contrast, a Huber function was implemented as a regularization term R(x,δ):

$$R(x,\delta )=\left\{\begin{array}{cc}-x^2,& \left|x\right|<\delta \\ 2\delta \left|x\right|-{\delta }^2,& \mathit{\text{else}}\end{array},\right.$$ (4)

where x is the difference between a target voxel and its neighboring voxels and δ is a threshold.

The objective function Φ is then defined as:

$$\mathrm{\Phi }=-L+\lambda R,$$ (5)

where λ is the regularization strength.

The Separable Quadratic Surrogate (SQS) method was used to iteratively approximate the objective function Φ with a parabolic surrogate with known minima, where the surrogate is updated at the solution of each iteration.

Initially, this algorithm was directly applied to the current multi-source geometry by stitching the reconstructed images from each LE/HE source pair[45]. However, this approach introduced inter-slice inconsistencies and did not fully utilize the projection data in overlapping regions.

In the present study, the algorithm was modified to accommodate the multi-source cone-beam geometry. First, the raw projections from each focal spot were cropped to the illuminated region determined by the cone-angle beam geometry of each source-detector pair. The cropped projections were then grouped into LE and HE channels based on the energy filter settings. For the one-step inversion, forward projections were performed sequentially for each focal spot within a given energy channel using its corresponding source-detector geometry. This ensures geometric consistency between the predicted and measured projections for each individual source position. The predicted projections from both energy channels were then compared with the measured data, and the material density maps were iteratively updated to minimize the discrepancy. Virtual monoenergetic images were synthesized from a linear combination of the materials’ mass attenuation coefficients and mass densities at an energy E, as given in Equation (2).

Figure 4 demonstrates the workflow of the proposed DE-MS-CBCT system.

Figure 4.

Workflow of the material decomposition and image reconstruction process utilizing the Adjacent Scatter Ratio Subtraction (ASRS) method and a one-step inversion material decomposition algorithm tailored for the DE-MS-CBCT system.

G. Image Analysis

The mass densities of the iodine and calcium inserts and the SolidWater were measured from the material density maps obtained from two basis material decomposition. For each insert, the average mass density and its standard deviation within each 20 mm diameter circle region of interest (ROI) across four central axial slices of 3 mm thick was calculated. For the water, the densities and standard deviations of 20 ROIs (five 20 mm diameter circles per axial slice over four 3 mm slices) were computed.

The accuracy of the iodine and calcium mass densities obtained from the material decomposition was assessed using the mean absolute error (MAE):

$$MAE=\frac{1}{n}\sum _{i=1}^n\left|{\rho }_i-\widehat{{\rho }_l}\right|,$$ (2)

where n is the number of inserts, ρ_i is the nominal density and $\widehat{{\rho }_l}$ is the measured density.

The cupping artifact was evaluated by analyzing the line profile of the central axial slices of the VMI. The degree of cupping artifact (t_cup) was quantified by computing:

$$t_{cup}=100\times \frac{{\mu }_{max}-{\mu }_{min}}{{\mu }_{max}},$$ (3)

where μ_max and μ_min are the maximum and minimum values of the measured linear attenuation coefficients, respectively.

The CNR of the iodine and calcium inserts against the SolidWater background was measured from the VMI of the phantom at different energies. The mean HU values over the selected 16 mm × 16 mm × 3 mm ROIs of the inserts were obtained. For the SolidWater background, to decouple the impact of the cupping effect, two 16 mm × 16 mm × 3 mm ROIs around each insert, having the same radius as the insert, were selected to calculate the averaged SolidWater background HU values and standard deviations. The CNR was calculated by:

$$CNR=\frac{\left|H{U}_{\mathit{\text{inserts}}}-H{U}_{\mathit{\text{SolidWater}}}\right|}{{\sigma }_{\mathit{\text{SolidWater}}}},$$ (4)

where HU_inserts and HU_SoliWater are the mean HU values over the ROIs, σ_SolidWater is the standard deviation over the ROI in the background.

III. Results

A. Air-Kerma Measurement

The X-ray profiles of the eight alternating beams from the DE-MS-CBCT measured at the detector surface without an object present are shown in Figure 5. The average air-kerma rate recorded at the detector surface was 721 μGy/s for the four LE beams and 407 μGy/s for the four HE beams. Minimal variation between consecutive exposures for each energy spectrum and negligible rising and falling times for each X-ray pulse were observed. The DAP value for the LE and HE beams were 394.0 mGy·cm² and 221.6 mGy·cm², respectively, leading to a combined DAP of 615.6 mGy·cm² for a full DE-MS-CBCT scan. The total imaging air-kerma at the system isocenter was 3.65 mGy, with 2.34 mGy for the LE scan and 1.31 mGy for the HE. This total imaging air-kerma for the DE-MS-CBCT is the same as a clinical single energy dental CBCT (Newtom 5G) for an average-size adult in the standard protocol in our institution. The total imaging dose for the DE-CBCT was 3.79 mGy (2.45 mGy LE and 1.34 mGy HE), similar to that of the DE-MS-CBCT. The results are listed in Table 1.

Figure 5.

The air-kerma rates of the eight X-ray beams of the DE-MS-CBCT measured at the detector surface without the object using imaging protocol of this study. The intervals between LE and HE pulses were set to accommodate the detector readout time.

Figure 6 shows the sagittal views of the 70 keV VMI and the corresponding line profiles of the Defrise phantom generated from the data collected using the DE-MS-CBCT and DE-CBCT. Severe cone beam artifacts in the form of image distortion and blur are observed in the images obtained using the conventional DE-CBCT configuration (Figure 6b). In the line profile drawn along the vertical direction, the distinction between the acrylic discs and the air gap becomes blurred when moving away from the central axial plane. These image artifacts are eliminated in the image from the DE-MS-CBCT. The line profiles of the peripheral and central acrylic discs are essentially the same. This is attributed to increased data sampling in the DE-MS-CBCT because of the reduced X-ray beam cone angle. The result is consistent with what was observed in the single-energy ms-CBCT[29].

Figure 6.

Sagittal views of the virtual monoenergetic images of the Defrise phantom (70 keV) from the DE-MS-CBCT (a) and DE-CBCT (b). (c, d) Corresponding axial line profiles measured in the ROIs highlighted by yellow boxes. Image window:−1000 HU to 200 HU. Both volumes were reconstructed with an isotropic voxel size of 1.0 mm.

C. Material Decomposition

Figure 7 shows the mass density maps and 70 keV VMI of the contrast phantom reconstructed with 1 mm axial resolution and 3 mm slice thickness, using the dataset from the DE-MS-CBCT and DE-CBCT. The water density maps show that the severe CBCT cupping artifacts (shadow regions) in the DE-CBCT images (e and k) are substantially reduced in the DE-MS-CBCT images (b and h).

Figure 7.

Axial images of the contrast phantom illustrating mass density maps obtained from two-material decomposition: iodine/water base (iodine maps: a, d; water maps: b, e) and their corresponding virtual monochromatic images (VMIs) at 70 keV (c, f); calcium/water base (calcium maps: g, j; water maps: h, k) and corresponding VMIs (i, l).

The mass densities of the inserts and water calculated from the selected ROIs in these images are listed in Table 2. The MAE was computed for iodine and calcium inserts using the nominal densities provided by the manufacturer as the reference. Comparing the results from the DE-MS-CBCT and DE-CBCT, the main difference is the significantly improved accuracy of the water mass density, from 854 ± 35 mg/ml by DE-CBCT to 984 ± 9 mg/ml by DE-MS-CBCT in the case of iodine/water phantom; and from 840 ± 41 to 982 ± 13 in the case of calcium/water phantom. The MAE for iodine and calcium densities were similar between the two scanning systems.

Table 2.

A Summary of Iodine Calcium and Water Mass Densities.

	Ground Truth (mg/ml)	DE-CBCT (Relative Error)	DE-MS-CBCT (Relative Error)
Iodine	15	14.64 ± 0.83 (2.40%)	14.71 ± 0.79 (1.93%)
	10	10.17 ± 0.72 (1.70%)	10.12 ± 0.81 (1.20%)
	5	5.48 ± 0.69 (9.60%)	4.78 ± 0.74 (4.40%)
	2	2.28 ± 0.56 (14.00%)	1.46 ± 0.66 (27.00%)
	MAE	0.32	0.27
Water	1000	854.27 ± 34.81 (14.57%)	983.66 ± 8.67 (1.63%)
Calcium	300	287.95 ± 18.50 (4.02%)	313.25 ± 15.88 (4.42%)
	100	99.93 ± 7.45 (0.07%)	99.77 ± 10.67 (0.23%)
	50	54.53 ± 7.18 (9.06%)	44.86 ± 9.88 (10.28%)
	MAE	5.55	6.21
Water	1000	840.12 ± 41.44 (15.99%)	981.67 ± 13.16 (1.83%)

D. Nonuniformity

The water density map’s uniformity was evaluated by calculating the standard deviation of the mean water mass density within 20 circular ROIs distributed across four central axial slices. Figure 8(a) presents the voxel-wise water density distributions obtained from the DE-MS-CBCT, DE-MS-CBCT without ASRS, and conventional DE-CBCT. Reducing the cone angle from 10° in DE-CBCT to 4° in DE-MS-CBCT without ASRS decreased the variations to 21.50 mg/mL (I/Water) and 25.84 mg/mL (Ca/Water), indicating the benefit of the smaller cone angle. With ASRS applied, DE-MS-CBCT further reduced the variations to 8.67 mg/mL (I/Water) and 13.16 mg/mL (Ca/Water). Figure 8(b) presents the line profiles of the 70keV VMI of the iodine/water phantom. The DE-CBCT image exhibited a substantial cupping artifact (t_cup = 14.53%). When the cone angle was reduced, t_cup decreased to 5.43%, and it was further reduced to 2.94% when ASRS was applied in DE-MS-CBCT.

Figure 8.

(a) Violin plots with superimposed box plots show the voxel-wise water-density distributions (mg/ml) in the DE-MS-CBCT, DE-MS-CBCT without ASRS, and conventional DE-CBCT images. Twenty circular ROIs, five in each of four central axial slices, were analyzed; their positions are marked by yellow circles in the inset image. (b) Line profiles of the normalized attenuation coefficient are extracted from a box-shaped ROI (yellow line) in the 70 keV VMI of the iodine/water phantom for DE-MS-CBCT, DE-MS-CBCT without ASRS, and DE-CBCT.

E. Contrast-Noise-Ratio

Figure 9 shows the CNRs for the 15 mg/ml Iodine and 300 mg/ml Calcium inserts against the SolidWater background measured from the VMIs of the contrast phantom in the energy range of 50 keV to 150 keV for DE-MS-CBCT, DE-MS-CBCT without ASRS, and DE-CBCT. The CNR of DE-MS-CBCT without ASRS is consistently higher than that of DE-CBCT, and the application of ASRS in DE-MS-CBCT further improves the CNR across all energy levels.

Figure 9.

The CNRs of the 15 mg/ml iodine (a) and 300 mg/ml calcium (b) inserts against the SolidWater background in the VMIs from 50 keV to 150 keV.

Table 3 lists the CNR and contrast for all seven inserts at 70 keV for the two scanners. DE-MS-CBCT significantly outperforms DE-CBCT across all tested concentrations. For the iodine inserts, the percentage improvement of the CNR ranges from 4.8% to 40.0%, while the contrast improvement ranges from 14.1% to 25.7%, depending on the insert density. For the calcium inserts, the CNR improvement ranges from 6.4% to 53.4%, with the improvement in contrast around 24%.

Table 3.

CNR and Contrast of Different GAMMEX Inserts Measured from VMIS at 70 keV, scanned by DE-CBCT and DE-MS-CBCT

GAMMEX Inserts		DE-CBCT		DE-MS-CBCT		Improvement
		CNR	Contrast (HU)	CNR	Contrast (HU)	CNR	Contrast (HU)
Iodine	15 mg/ml	18.0	340	23.0	388	27.8%	14.1%
	10 mg/ml	12.1	216	13.0	257	7.4%	19.0%
	5 mg/ml	6.2	108	6.5	125	4.8%	15.7%
	2 mg/ml	2.0	35	2.8	44	40.0%	25.7%
Calcium	300 mg/ml	30.9	715	47.4	877	53.4%	22.7%
	100 mg/ml	14.1	242	15.0	300	6.4%	24.0%
	50 mg/ml	7.5	139	9.5	175	26.7%	25.9%

F. Anthropomorphic Head Phantom

Figure 10 compares the calcium- and water-density maps derived from calcium–water material decomposition, together with the corresponding 70 keV VMIs, for the DE-CBCT and DE-MS-CBCT systems. Consistent with the observations in Figure 6, the DE-MS-CBCT VMIs exhibit fewer shading artifacts and a larger effective FOV than the DE-CBCT VMIs. In the calcium-density maps in Figure 10 (a) and (b), both systems clearly demonstrate the high-density bone and dental structures. The 70 keV VMIs further show that the diamondshaped shading artifact at the center and the artifacts adjacent to the dental structures are visibly weaker in the DE-MS-CBCT image in Figure 10(f) than in the DE-CBCT image in Figure 10(e). In addition, Figure 10(g) and (h) demonstrate an overall reduction in shading artifacts and an enlarged effective FOV achieved with DE-MS-CBCT.

Figure 10.

Images of the anthropomorphic head phantom. Calcium- and water-density maps from the conventional DE-CBCT (a, c) and the DE-MS-CBCT (b, d). Axial and sagittal slices of the 70keV VMIs from the conventional DE-CBCT (e, g) and the DE-MS-CBCT (f, h). The voxel size is 0.8(x) × 0.8(y) × 1.2(z) mm³. The bright small cylindrical object is a titanium dental implant.

Figure 11 presents the HU line profiles extracted from the yellow boxed ROI in the 70 keV VMIs of the RANDO head phantom. Compared with DE-CBCT, DE-MS-CBCT without ASRS already yields a flatter profile in the soft tissue equivalent region, and DE-MS-CBCT with ASRS further improves the uniformity. The diminished central hump from DE-CBCT is attributed to the scatter reduction from both cone-angle reduction and the additional scatter removal provided by ASRS[28].

Figure 11.

Line profiles extracted from the axial view slices of the RANDO head phantom VMIs at 70 keV.

IV. Discussion

This study demonstrates a method to perform dual energy CBCT using a distributed X-ray source array, interlaced source spectral filtration, and one-step image reconstruction and decomposition algorithm, without either kV switching or an energy sensitive FPD. Phantom imaging results show that, compared to a reference DE-CBCT, it reduces scatter and cone-beam artifacts and consequently improves image uniformity and CNR without increasing the imaging dose. These improvements result from significantly reduced cone angle of each X-ray beam (4.1° degrees) compared to a conventional single source CBCT (10.5° degrees for the same geometry). These results are consistent with the findings from previous studies of the single energy ms-CBCT[28, 29, 31], which has shown significantly reduced scatter-primary-ratio, HU nonuniformity and cone-beam artifact compared to a conventional CBCT.

Because of the improved image spatial uniformity, the DE-MS-CBCT enhances the water mass density accuracy derived from materials decomposition, compared to the reference DE-CBCT using two wide cone angle X-ray beams with different energies[20]. The mean relative error (averaged over the iodine/water and calcium/water images) is reduced from 15.3% in DE-CBCT to 1.7% for DE-MS-CBCT, which is within the range of the reported values for fast kVp switching (4.7%) and dual-layer detector (2.8%) based DE-CBCT after applying software-based scatter correction.[46] The accuracy obtained for iodine and calcium inserts is comparable to that of the reference DE-CBCT, and previously reported results using dual-layer detector[11] and kVp switching[47] based DE-CBCT after scatter correction.

In this study, the imaging air-kerma of the DE-MS-CBCT was selected to match that of a standard single-energy dental CBCT protocol for an adult patient in our institution, and the air-kerma of the reference DE-CBCT to match that of the DE-MS-CBCT. The partial overlap of the X-ray beams within each energy group (Fig. 1a) was taken into account in the DAP and mean air-kerma calculations. Because of the difference in the beam overlap area, the X-ray exposure of the DE-MS-CBCT is lower than that of the DE-CBCT for roughly the same air-kerma. A limitation of this study is that these air-kerma values were derived from DAP measurements performed in air at the detector surface without a phantom. A comprehensive patient-dosimetry assessment would provide additional validation and is planned for future clinical translation studies.

The image artifacts observed near the center of the VMIs in Figure 7 and Figure 10 are attributed to the detector offset geometry (to extend the field of view) used in this study. Similar artifacts were also present in the reconstructed images from the study using the offset detector CBCT[48, 49], including our previous single-energy ms-CBCT device[28, 50], and this artifacts could be mitigated by applying a weighting factor in the either image domain[48, 50] or projection domain[49]. In the current dual-energy dataset, no additional processing was applied.

The voxel intensity line profile in Figure 11 shows a marked reduction in the central hump from DE-CBCT to DE-MS-CBCT, primarily due to scatter reduction, consistent with the central artifact reduction observed in Figure 10(c, d). A similar trend was reported in our previous single-energy ms-CBCT study[28], which exhibited more pronounced cupping artifacts; in that case, the central hump was mitigated by decreasing the cone angle and further reduced through the application of ASRS.

In this study, the same spectral filters in our previous study[20] were used. The low- and high-energy beams generated have a respective mean energy of 60.4 keV and 75.9 keV at the applied X-ray tube voltage of 110 kVp. The spectral separation is similar to the reported values for a commercial source filtration-based DE-CT[14], and fast kVp switching and dual-layer detector-based DE-CBCT[11, 12, 46]. Further increasing the spectral separation is, in principle, possible by using stronger filtration, but at the expense of a reduced air-kerma rate.

The multi-source approach for DE-CBCT significantly increases the number of image frames required, which is proportional to the number of X-ray sources used. This increases the image acquisition time. With each beam from the fixed-anode CNT X-ray source array operating at the output level comparable to slightly high than a convention fixed-anode dental X-ray tube at similar focal spot size, a DE-MS-CBCT scan using the protocol of this study takes 46.7s to acquire the dataset. This is to be compared to the reported value of over 30s for a kVp switching based dental DE-CBCT[51]. The dominant contribution to the increased imaging time is the detector readout (32.3s) instead of the X-ray exposure time (14.4s). The prolonged imaging time can potentially be mitigated by several ways. (1) The current imaging protocol uses 360 projection views (1440 image frames) per energy. With rapid advances in image processing and machine learning techniques, the number of project views can potentially be reduced without sacrificing the image quality to speed up the data acquisition. (2) The current benchtop device uses a half-width FPD and an offset geometry to cover the entire FOV. Replacing it with a full-width FPD operating at the frame rate afforded by the current technology can bring the imaging time to within the range of the current clinical dental CBCT scanners. (3) Implementing a ROI detector readout strategy can substantially reduce the detector readout time. In the current detector configuration, the readout time is proportional to the number of detector rows being read. In multisource CBCT imaging, only a small portion of the detector is exposed during each X-ray pulse due to the spatially distributed source geometry. By selectively reading out only the exposed detector region rather than the entire detector area, the readout time can be significantly reduced.

Although the current study focused on two material decomposition, the proposed multisource architecture could be extended to multi-energy configurations for multi-material decomposition[52, 53], as well as other clinically relevant applications such as direct volumetric bone mineral density (BMD) quantification[54] for dental implant planning, quantitative electron density mapping[55], and stopping-power ratio estimation[56] for radiotherapy. These applications require further investigation and will be explored in future work.

V. Conclusion

This study presents an approach for achieving high-quality DE-CBCT imaging by utilizing multiple axially collimated X-ray beams produced by a CNT X-ray source array with interlaced spectral filters, without using kVp switching or an energy-sensitive detector. Compared to the conventional methods for DE-CBCT, it eliminates cone beam artifacts, reduces the spatial nonuniformity, improves the accuracy of the water mass density, and enhances the image contrast.