Effect of scatter suppression with 2D antiscatter grids in photon counting compact CBCT

Abstract

Energy sensitive and photon counting detectors can provide improved tissue visualization and material quantification capabilities in Cone Beam Computed Tomography (CBCT) systems. However, their implementation in CBCT systems is more challenging, which is in part due to high fluence of scattered X-rays in wide cone angle CBCT geometry. Specifically, high scatter contamination in lower energy spectrum challenges reconstruction of high fidelity CBCT images by using lower energy X-rays.

To address this problem, we investigated a robust scatter rejection with 2D antiscatter grids in a benchtop photon counting and compact CBCT system. The benchtop system employs a 35 cm wide CdTe photon counting detector with two energy thresholds. To reject scatter, a dedicated 2D antiscatter grid (2D grid) prototype made from tungsten was developed and mounted directly on the detector. To correct residual scatter not stopped by the 2D grid, a measurement-based scatter correction method, referred to as Grid-based Scatter Sampling (GSS), was utilized.

Without 2D grid, scatter to primary ratio (SPR) reached 2.3 in the 15–40 keV energy bin. SPR was factor of 3 higher in the lowest energy bin when compared to the highest energy bin (90–120 keV). With the 2D grid, SPR was reduced below 0.14, and SPR values were more homogenous across the energy spectrum. CT number nonuniformity was factor of 3 lower in both low and high energy bin CBCT reconstructions. Improvement in contrast to noise ratio and contrast was more pronounced in the low energy bin CBCT images.

This work indicates that 2D grids can significantly reduce spectral contamination caused by scatter in photon counting compact CBCT, and potentially enable higher fidelity CBCT image reconstructions.

1. INTRODUCTION

Photon counting detectors have potential advantages over standard, energy integrating flat panel detectors in CBCT imaging, such as extraction of energy information from X-rays, immunity to electronic noise, higher spatial resolution, and higher quantum efficiency¹. However, in CBCT systems with large cone angles, scattered radiation biases X-ray counts in energy bins of a photon counting detector, degrading quantitative accuracy and image quality in photon counting CBCT images. Moreover, increased X-ray fluence due to scatter may potentially contribute to pulse pile-up and limit count rate capability. Even though radiographic 1D antiscatter grids can help to reduce scatter fluence, a nonnegligible percentage of scatter is still transmitted through the grid, contaminating the counts in energy bins². As an alternative to radiographic 1D antiscatter grids, 2D antiscatter grids reduce scatter further, and may minimize X-ray count bias caused by scatter^{3, 4}.

To investigate the feasibility of 2D antiscatter grid in photon counting CBCT, a benchtop photon counting CBCT system was developed. The benchtop CBCT system emulated compact CBCT system geometry with short source to detector distance, typical in point-of-care portable CBCT systems used in neurotrauma, extremity, and dental imaging^{5, 6}. A 2D antiscatter grid prototype was developed and installed in the system. 2D grid’s effect on energy spectrum and CBCT image quality was evaluated in a series of experiments. Since large area photon counting detectors suitable for CBCT are still under development, a line photon counting detector was employed in this work.

2. MATERIALS AND METHODS

Photon counting compact CBCT was emulated using a benchtop X-ray imaging system. This system employs a Dunlee DA1094 X-ray tube, EMD Technologies EPS 80 generator, and a Parker Motion direct drive rotary stage (Figure 1). Images were captured using a Varex TDI 350 photon counting detector with a 350×6 mm² active area, 0.75 mm thick Cadmium Telluride (CdTe) sensor, with 100 μm pixels⁷. All projections were acquired with anticoincidence circuitry turned on. The X-ray source to detector distance was 75 cm, source to isocenter distance was 50 cm.

Figure 1.

(a), Compact photon counting CBCT setup. (b), 2D antiscatter grid mounted on the detector.

CBCT scans were acquired with a field of view (FOV) of 16 cm in the axial direction at isocenter for all acquisitions. Projection data was acquired using two energy bins: Low energy bin (LE) = 25–60 keV, High energy bin (HE) = 60–120 keV. In addition, images were reconstructed by summing both energy bins, (TE) = 25–120 keV. Scans were performed using continuous fluoroscopy X-ray acquisition. A total of 576 projections were acquired for each scan. Scans were taken using 120 mAs, 120kVp, resulting in CTDI value of 30.5 mGy. Projections were reconstructed using filtered backprojection, and 0.4×0.4×0.4 mm³ voxel size.

The 2D grid prototype was additively manufactured using pure tungsten and powder bed laser melting process. The grid was comprised of an array of square holes, all aligned to the focal spot (Figure 2). The 2D grid prototype has 1.6 mm grid pitch, 0.1 mm wall thickness, and a grid ratio of 12. The overall grid size is 400 × 27 mm², comprised of two 200 mm length modules.

Figure 2.

Focused 2D grid CAD model and the fabricated prototype.

Since a small percentage of scattered X-rays pass through the 2D grid, remaining scatter in projections were corrected by using the Grid-based Scatter Sampling (GSS) method^{8, 9}. GSS method employs the 2D grid itself as a scatter measurement device; 2D grid’s wall shadows introduce spatially periodic intensity variations across the active area of the 2D grid. This intensity variation, i.e. the ratio of photon counts in the grid holes to grid shadows, is proportional to the residual scatter content. Thus, this relationship between signal intensity variation and scatter intensity was exploited to measure the residual scatter intensity.

Scatter rejection performance of the 2D grid was measured as a function of energy bins and FOV in the axial direction. Scatter intensity was measured using the beam stop method. An acrylic cylinder with a diameter of 20 cm was placed at the isocenter to generate scatter. A 6 mm thick tungsten bar was used as a beam stop and placed in front of the X-ray source. Scatter intensity, $S$, was measured in the beam stop’s shadow, along the entire width of the acrylic cylinder. Total X-ray intensity, $T$, consisting of scatter and primary was measured in the same locations, but without the beam stop. $\mathrm{S}\mathrm{P}\mathrm{R}$ was calculated using the following equation:

$$SPR=\frac{S}{T-S}$$ #(1)

SPR was measured at 2, 7, and 13 cm FOV in axial direction at isocenter. In addition, scatter transmission fraction was calculated from the ratio of scatter intensities measured with and without the 2D grid.

HU nonuniformity was calculated in the Catphan 504 material insert module. Regions of interest (ROIs) were selected in uniform regions throughout the area of the phantom, and HU values were averaged in each ROI. HU nonuniformity was calculated as follows,

$$HUnonuniformity\left(i\right)=HU\left(i\right)-mean\left(HU\right)$$ #(2)

Where $HU\left(i\right)$ is the mean value of i^th ROI and $mean\left(HU\right)$ is the ensemble average of mean $\mathrm{H}\mathrm{U}$ values.

Contrast to noise ratio ($\mathrm{C}\mathrm{N}\mathrm{R}$) was calculated for the PTFE (Teflon) and Delrin material inserts in the phantom. CNR is defined as

$$CNR=\frac{I\left(object\right)-I\left(background\right)}{\frac{{\sigma }_{object}+{\sigma }_{background}}{2}}$$ #(3)

Where $I$ is the mean HU value in the object of interest and the uniform surrounding region representing the background. $\mathrm{\sigma }$ is the standard deviation of HU values in the object and the background.

3. RESULTS

The primary transmission of the 2D grid was measured using two energy bins, LE (25–60 keV), HE (60–120keV), and primary transmission histograms were generated (Figure. 3). The average primary transmission was 80% in all energy bins, and the primary transmission was uniform across the active area of the 2D grid. 50% of the pixels received 97% or more primary transmission.

Figure 3.

(a), Average primary transmission of the 2D grid as function of ROI location (One sample ROI is shown by red rectangle). (b), primary transmission histogram (PTH) of one of the ROIs.

SPR across 20 cm diameter phantom and in four energy bins are shown Figure 4. In no grid images, SPR reached 2.3 in the lowest energy bin (15–40 keV) and at 13 cm axial collimation. SPR was a factor of 3.3 higher in the lowest energy bin than the highest energy bin (90–120 keV), implying the challenges in high fidelity CBCT reconstruction at lower x-ray energies using photon counting detectors. The 2D grid reduced SPR from 2.3 to below 0.14 in the 15–40 keV energy bin and made SPR values more comparable across different energy bins. Similar trends were observed for all collimation widths in the axial direction.

Figure 4.

Scatter-to-primary ratio for four energy bins as a function of x-ray collimation in the axial direction.

Scatter transmission fraction (Ts) of the 2D grid was below 10% at the thickest section of the phantom at 13 cm collimation in the axial direction (Figure.5). Scatter transmission increased as a function of energy, which is due to the more forward directed distribution of scattered X-rays at higher energies. Likewise, scatter transmission increased for smaller collimation widths in the axial direction due to more forward directed distribution of scattered X-rays reaching the 2D grid.

Figure 5.

Scatter transmission fraction (T_S) of the 2D grid prototype for four energy bins as a function of x-ray collimation in the axial direction.

CBCT images of the Catphan material inserts and the line pair modules acquired using 25–60 keV (LE) and 60–120 keV (HE) energy bins are shown in Figures 6 and 7, respectively. Without the 2D grid, loss of contrast and increased shading artifacts are evident in both LE and HE bins. Qualitatively, artifacts were more emphasized in the LE bin due to higher SPR at lower energies. With the 2D grid, shading artifacts were suppressed and contrast increased. Median HU nonuniformity was reduced from 15 to 3 HU for the HE Bin and from 18 to 2 HU for the LE Bin (Figure 8). For the PTFE material, the contrast was 307 and 363 HU for the LE and HE bins, respectively, without the grid (Figure 9). With the 2D grid, contrast increased to 790 and 725 HU in LE and HE energy bins, respectively. Similar trends were observed for other material inserts. As a result, CNR also increased by a factor of 1.51and 1.08 in LE and HE bins, respectively, with the 2D grid. The increase in CNR was more emphasized in the LE bin. This was partially attributed to larger scatter fraction in the LE bin without the 2D grid, and thus, scatter suppression provided a larger increase in CNR. HE images appeared noisier than the LE images, since counts were about 50% less in the HE bin. Spatial resolution was comparable in no-grid and with 2D grid images.

Figure 6.

CBCT images without 2D grid (top row) and with 2D grid (bottom row) at different energy bins. (HU window [−600 600])

Figure 7.

Line pair phantom images acquired with and without the 2D grid. (HU window [−200 2000]).

Figure 8.

(a), HU nonuniformity and (b), CNR with and without 2D grid, and as a function of energy bins. Both HU nonuniformity and CNR were improved in all energy bins with the 2D grid.

Figure 9.

Contrast measured (a), at PTFE (Teflon) and (b), Delrin inserts. With 2D grid: More than factor of 2 increase in contrast in both LE and HE energy bins when compared to “without grid”

4. SUMMARY

This preliminary investigation demonstrated the scatter rejection performance of a 2D antiscatter grid in the context of photon counting compact CBCT. The 2D grid provided more than an order of magnitude reduction in SPR in the 15–40 keV energy bin, indicating that the SPR reduction benefit of the 2D grid at lower X-ray energies. Correction of residual scatter by the GSS method in a photon counting detector was also demonstrated for the first time, which further improves CT number accuracy. With the 2D grid and the GSS method, HU nonuniformity was reduced by a factor of 3, and contrast was improved by a factor of 2 in both low and high energy bins. Conventional antiscatter grids may not improve CNR when imaging relatively small phantoms due to relatively low scatter fluence generated by smaller phantoms. This is mostly due to attenuation of primary X-rays and relatively low scatter suppression efficiency of conventional antiscatter grids. However, in this work, 2D grid provided significant improvement in CNR due to less than 10% scatter transmission at large X-ray cone angles and 80% primary transmission. Since lower energy bins have higher SPR values, CNR improvement by the 2D grid is larger at lower X-ray energies. Besides improved image quality, an order of magnitude reduction in SPR by the 2D grid reduces the X-ray fluence incident on the detector. In return, lower X-ray fluence may reduce pulse pile-up and count rate limitations of photon counting detectors. Benefits of 2D grid in reducing pulse pile-up remains to be investigated in future studies.