Megavoltage cross-scatter rejection and correction using 2D antiscatter grids in kilovoltage CBCT imaging

Abstract

Simultaneous use of kilovoltage (kV) and megavoltage (MV) beams has numerous potential applications in cone beam computed tomography (CBCT)-guided radiotherapy, such as fast MV+kV CBCT for single breath-hold scan, tumor localization with kV CBCT imaging during MV therapy delivery, and metal artifact suppression. However, the introduction of MV beams results in a large MV-cross scatter fluence incident on the kV Flat Panel Detector (FPD), and thus, deteriorating the low contrast visualization and Hounsfield Unit (HU) accuracy. In this work, we introduced a novel and robust method for reducing the effects of MV cross scatter. First, we implemented a 2D antiscatter grid atop the detector which rejects a large section of MV cross scatter. This hardware-based approach, while effective, allows a fraction of MV cross scatter to be transmitted to the FPD, resulting in artifacts and degraded HU accuracy in CBCT images. We thus introduced a data correction step, which aimed to estimate and correct the remaining MV cross scatter. This approach, referred to as Grid-Based Scatter Sampling, utilized 2D antiscatter grid itself to measure and correct remaining MV cross scatter in projections. We investigated the performance of the proposed approach in experiments by simultaneously acquiring kV CBCT and delivering MV beams with a clinical linac. The results show that the proposed method can substantially reduce HU inaccuracy and increase contrast-to-noise ratio (CNR). Our method does not require synchronization of kV and MV beam pulses, reduction of kV frame acquisition rate, or MV dose rate, and therefore, it is more practical to implement in radiation therapy clinical setting.

1. INTRODUCTION

The concept of using MV and kV beams simultaneously in CBCT guided radiation has become an area of interest in recent years. One of the benefits of this approach is in reducing the CBCT scan time, by acquiring CBCT projections using both kV and MV imagers. It has been shown [1], [2] that such an MV+kV imaging scheme could reduce the imaging time to a half of kV only imaging time which could enable single breath hold CBCT scans to reduce breathing motion induced artifacts [3]. Furthermore, MV+kV implementation could be used for simultaneous imaging during treatment delivery as a means of ensuring tumor localization accuracy during radiation therapy [4] [5] [6]. This approach may allow calculation of delivered radiation therapy dose based on the anatomy at treatment delivery [7]. This implementation could also allow more precise target localization, which may alleviate any issues caused by organ location changes between setup and irradiation [8], allowing more accurate delivery of dose to tumors, and spare healthy tissues in single fraction or hypofractioned radiation therapy regimens such as Stereotactic Body Radiation Therapy (SBRT).

However, simultaneous MV+kV beam delivery results in a large MV-cross scatter fluence incident on the Flat Panel Detector (FPD) which deteriorates the quality of kV CBCT images. An example of this effect is presented in Fig. 1. Several solutions have been proposed to address this issue and suppress the effects of MV-cross scatter. Ling et al. [4] suggested the use of control points in Varian linacs for sequential triggering of kV and MV beams. With control points, two actions take place; the first is the MV delivery and the second is the kV projection acquisition while the MV beam is turned off to avoid MV-cross scatter. While this approach achieves MV cross scatter-free kV projections, the required synchronization between the processes for MV-radiation delivery and kV image acquisition may lead to MV beam delivery and kV projection acquisition interruptions. Van Herk et al. [3] on the other hand, suggested the use of an alternating pulse sequence, whereby the even frames would contain a kV image combined with MV scatter and the odd frames would only contain MV scatter while no kV pulse is generated. They would then subtract the MV-scatter only projections from those of MV+kV and reconstruct the result which would correct MV cross-scatter intensity. The drawback of this method is that it halves the frame rate of the kV detector and that noise due to MV cross-scatter remains in kV projections. Thus far, such scatter-correction methods focused on developing a method to break down the MV+kV delivery into alternating sequences of MV+kV and MV only or kV only irradiations within the same acquisition. As an alternative approach, Boylan et al. [7] suggested acquiring two sets of CBCT scans during radiation therapy delivery, one concurrent with MV beam delivery with kV beam off to measure patient specific MV cross-scatter. In the subsequent CBCT scan with kV beam on and concurrent MV beam delivery, MV cross scatter is corrected using the data from the prior scan. However, this approach requires patient anatomy and position to be the same in subsequent treatment sessions. Finally, Iramina et al. [5] suggested developing an empirical correction method, where MV cross scatter distributions were characterized using a phantom and changing the MV irradiation parameters. They would then use this model to estimate the resulting MV cross-scatter in different MV+kV delivery scenarios and correct them accordingly. However, using such a model-based method may not accurately estimate the MV cross scatter distributions due to discrepancy between patient-specific clinical imaging conditions, versus model data acquisition conditions.

Fig. 1:

Example of image quality deterioration due MV cross scatter fluence incident on the kV FPD due to simultaneous MV beam delivery. CBCT scans were acquired using a standard 1D antiscatter grid. HU window range: [−500 0]

In this paper, we proposed a 2D antiscatter grid and a scatter-correction method to mitigate the effects of MV cross scatter in kV CBCT images. Our method used only the original MV+kV projections as the input, nor did it require kV and MV beam sequencing, and thus, it did not change the image acquisition rate. This method exploited the use of a 2D antiscatter grid prototype mounted on the FPD which has been shown to improve kV scatter suppression and CBCT image quality when compared to high-performance radiographic grids [9–11]. Furthermore, we implemented the Grid-based Scatter Sampling (GSS) method [12, 13] to remove the remaining MV cross-scatter that is not rejected by the 2D antiscatter grid. This method used 2D grid itself as a residual scatter measurement device.

2. MATERIALS AND METHODS

Experiment Setup:

CBCT experiments were performed with a Varian TrueBeam CBCT in offset detector geometry (i.e., half fan geometry). For each kV CBCT acquisition, 900 projections were acquired at 60mA and 20 mSec per projection at 125 kVp for head sized and pelvis sized phantoms. A 0.9 mm titanium beam filter and bowtie filter were used. kV CBCT imaging during SBRT delivery was emulated by using the 6 MV flattening filter free beam, setting the MV field size to 3×3 cm² and delivering 1200 Monitor Units (MU) at a rate of 1200 MU/min during kV CBCT acquisition, corresponding to 1.33 MU per kV projection.

An initial set of kV and MV+kV projections using only the default TrueBeam grid (Smit Rontgen 1D grid, grid ratio: 10) was acquired for comparison. Then, a focused, tungsten 2D antiscatter grid prototype with a grid ratio of 12, grid pitch of 2 mm, and septal thickness of 0.1 mm was installed on the FPD (As pictured in Fig. 2). Properties and the imaging performance of the grid was described in prior publications[14]. A second set of kV and MV+kV projections using this new 2D grid were then acquired.

Fig. 2:

2D Antiscatter Grid as utilized for experiments

CBCT images were reconstructed by using FDK method at 0.9×0.9×1.0 mm³ voxel size and with a 3×3 pixel binning in projections [15]. The image noise and HU values were calculated in multiple regions of interest to quantify the performance of our method.

Grid-Based Scatter Sampling (GSS) Method:

In the next step, we corrected the residual MV cross scatter using the GSS method. In GSS, 2D grid’s septa act as a micro-array of beam modulators placed on the detector, introducing a periodic signal intensity variation across grid’s septa and holes in projections [9] [12]. On the other hand, when residual scatter intensity, S, is present in projections, signal intensity pattern changes as a function of S. Thus, assuming S is approximately uniform in pixels residing both in grid shadows and grid holes in a small neighborhood of pixels, S can be calculated as in following,

$$S(x,y)=\frac{d(x,y)}{{GM}_{grid}\left(x,y\right)-{GM}_{hole}(x,y)}$$

where d is the signal difference in grid shadows and holes in a small neighborhood, GM_grid and GM_hole are the values of gain maps in grid septal shadows and holes, respectively. Gain maps are calculated from flood projections using the following

$$GM(x,y)=\frac{C}{F(x,y)}$$

where, F (x, y), is the flood projection, and C is the normalization constant. Using the above formulations, we could thus estimate scatter in grid shadows and use interpolation to find scatter values in grid holes. Finally, the calculated scatter is subtracted from projections to achieve scatter corrected projections. This approach simultaneously corrects both residual kV scatter and MV cross-scatter bias in kV projections.

Measures of Comparison:

1. Qualitative evaluations:

MV+kV CBCT image quality in the presence of 1D grid, 2D grid and 2D grid with GSS were evaluated visually.

2. HU Loss:

HU degradation in MV+kV CBCT images were quantified with respect to kV only CBCT images. We evaluated 1D grid, 2D grid, and 2D grid+GSS configurations. Specifically, we introduced ΔHU_method which represents the average HU difference between any method’s output for MV+kV and their output for their kV only projections. Calculation was as follows:

$${\mathrm{\Delta }HU}_{method}=|{HU}_{kVOnly,method}-{HU}_{kV+MV,method}|$$

We calculated average HU difference for a large region of interest encompassing a transverse CBCT image slice.

3. Contrast-to-noise Ratio (CNR):

We looked at specific regions of interest and formulated the contrast-to-noise ratio as follows:

$$CNR=\frac{mean\left({HU}_{ROI}\right)-mean\left({HU}_{background}\right)}{{(\sigma }_{ROI}+{\sigma }_{background})/2}$$

Where $mean\left({HU}_x\right)$ and ${\sigma }_x$ represent the average HU value and the standard deviation in a neighborhood respectively. Furthermore, ROI represents an area within each contrast object and background indicates the area surrounding the contrast object.

4. HU Profiles:

We qualitatively evaluated the HU profiles in each reconstruction and investigate how the addition of a 2D grid and 2D grid with GSS could affect the profile.

3. RESULTS AND DISCUSSION

3.1. CBCT reconstructions

Fig. 3(a) shows the images of the head sized phantom in both kV only and MV+kV scans with 1D grid, and their difference. As can be seen, HU degradation and image artifacts are noticeable due to MV cross scatter. HU degradation is more pronounced in the image acquired with 1D grid and for bone like objects due to locally higher MV cross scatter to primary ratio.

Fig. 3:

Reconstruction of head sized phantom using (a)1D Grid (b)2D Grid and (c)2D Grid + GSS scatter correction in the presence of BowTie Filter; window levels are [−250 250] for No MV and With MV and [−100 300] for Difference

Fig. 3(b) shows the head sized phantom under the same irradiation conditions as above, but by using the 2D grid. 2D grid rejects a significant amount of MV cross-scatter from the projections, and as a result HU degradation is less pronounced.

Finally, Fig. 3(c) utilizes the same raw data set in Fig. 3(b), but with the addition of GSS-based scatter correction. As observed, the difference between kV only and MV+kV images is quite small. Thus, the use of 2D grid with an additional GSS-based scatter correction could effectively suppress the effects of MV cross scatter for the head sized phantom.

Fig. 4 shows a similar experiment but with a pelvis sized phantom. Similar to the observations in head sized phantom images, employing 2D grid improves the image quality in comparison with the default 1D grid while implementing 2D grid with GSS-based scatter correction yields better image quality.

Fig. 4:

Reconstruction of the pelvis sized phantom using (a)1D Grid (b)2D Grid and (c)2D Grid + GSS scatter correction; window levels are [−450 250] for No MV and With MV and [−50 200] for Difference

3.2. HU Loss

Fig. 5 shows a boxplot for HU loss within phantoms when switching from kV only irradiation to MV+kV irradiation. In the case of head sized phantom, an average HU loss of 19.1±23.2 was observed for the 1D grid. By using the 2D grid, this loss was reduced to 9.8±12.2 while the addition of GSS-based scatter correction further reduced HU loss to 5.4±4.4. Pelvis sized phantom, in general, suffers larger HU loss due to a larger size which results in a reduction of primary and thus a higher Scatter to Primary Ratio (SPR). Specifically, in the case of pelvis sized phantom, an average HU loss of 107.3±53.6 was observed for the 1D grid. By using the 2D grid, this loss was reduced to 74.5±51.7 while the addition of GSS-based scatter correction further reduced HU loss to 40.4±31.6.

Fig. 5:

HU Loss Boxplot for Two Protocols using different Grid Types and Scatter Correction Methods and the corresponding ROI for this comparison

In both scenarios, using 2D grid in conjunction with GSS-based scatter correction provided the lowest HU loss (an overall HU loss drop-off factor of 3.1).

3.3. Contrast-to-noise Ratio (CNR)

Fig. 6 shows the CNR box plot for 8 contrast objects for (a) head sized phantom and (b) pelvis sized phantom respectively. Each box plot consists of one box and two whiskers where the line within the box is median; upper and lower edges of the box correspond to 75th and 25th percentile and upper and lower whiskers correspond to maximum and minimum data points. For the head sized phantom -where MV cross scatter fraction is less due to a smaller size- MV cross scatter results in a slight degradation of CNR most notable in 1D grid. Effect of MV cross scatter on CNR was less apparent for 2D grid and 2D grid+GSS. For the pelvis sized phantom, MV cross-scatter fraction was larger in the projections due to reduced primary intensity. As a result, CNR degradation due to MV cross scatter was more pronounced. When compared to 1D grid, 2D grid helped to increase the median CNR from 6.4 to 9.5. The use of GSS increased median CNR further, to 12.3. kV only CBCT images still provided high CNR than MV+kV CBCT images in all scatter suppression configurations, due to presence of relatively higher noise associated with MV cross scatter.

Fig. 6:

CNR boxplots for (a) Head sized Phantom and (b) Pelvis Sized Phantom using different grids and scatter correction methods

3.4. HU Profile

Fig. 7 shows the HU profiles of head sized phantom for the three scatter suppression methods. With 1D grid, a maximum reduction of 89.6 in HU values was observed due to MV+kV irradiation (continuous red line). When 2D grid was used, difference between kV only and MV+kV HU profiles was reduced with a maximum reduction of 32.7, as expected. Using 2D grid with GSS method further minimized this HU profile difference with a maximum reduction of 11.8.

Fig. 7:

HU profile of Head Sized Phantom along the dotted line shown in the inset.

Finally, HU profiles of the pelvis sized phantom are shown in Fig. 8. Due to increase in MV cross scatter fraction, a larger deviation (maximum of 270.4 HU) between kV only and MV+kV HU profiles was observed when 1D grid was in use. The 2D grid and addition of the GSS method reduced the maximum HU difference to 138.8, and 55.8 respectively.

Fig. 8:

HU profiles of the pelvis sized phantom along the dotted line indicated in the inset.

4. CONCLUSION

In this work, we introduced a novel and a robust method for rejecting and removing MV cross scatter from kV projections. We first reject a significant amount of scatter by implementing a 2D antiscatter grid. Due to 2D grid structure and 100 microns thick tungsten walls, 2D grid is more effective in rejecting MV cross scatter, which has average energy of 300 keV [16]. However, some of the MV cross scatter is still transmitted through the 2D antiscatter grid which degrades image quality, as demonstrated in our results. We then used our method of Grid-based Scatter Sampling (GSS) to further remove residual MV cross scatter. In experiments with head and pelvis sized phantoms and using a MV beam delivery scenario mimicking SBRT treatments, we demonstrated that our method effectively reduced HU inaccuracy and improved CNR. Specifically, we reduced average HU loss by a factor of 3.1 and increased retrieved median CNR in MV+kV irradiation by a factor of 2.3.

An important difference between the experiments performed with head and pelvis sized phantoms is the kV primary fluence. Pelvis sized phantom attenuates the kV primary beam more, resulting in higher MV cross scatter to primary ratio and larger degradation in CBCT image quality. This issue can be partially mitigated, by using higher kV imaging dose.

While the use of 2D grid reduced MV cross scatter intensity, new ring artifacts were introduced in CBCT images. This is caused by residual MV cross scatter in projections, resulting in suboptimal flat field correction and suppression of 2D grid septal shadows. Once the residual MV cross scatter was corrected by the GSS method, such ring artifacts were suppressed to a large extent. The correlation between ring artifacts and the residual scatter was discussed in detail in our prior publications[12, 13].

This study was performed by mimicking CBCT imaging conditions during MV radiation therapy, where a small field size and high dose rate for MV beam delivery was used. Alternatively, a large field size coupled with low dose rate MV beam delivery can be utilized to mimic simultaneous kV and MV projection acquisition in single breath hold CBCT. This approach remains to be investigated in future experiments.

Our evaluations were performed in the context of CBCT guided radiation therapy, proposed methods can be potentially used in other CBCT applications and beyond medical imaging, where high energy x-ray beams are employed, such as nondestructive testing and security imaging.