Improved image quality of cone beam CT scans for radiotherapy image guidance using fiber-interspaced antiscatter grid Uros Stankovic, Marcel van Herk, Lennert S. Ploeger, and Jan-Jakob Sonke Citation: Medical Physics 41, 061910 (2014); doi: 10.1118/1.4875978 View online: http://dx.doi.org/10.1118/1.4875978 View Table of Contents: http://scitation.aip.org/content/aapm/journal/medphys/41/6?ver=pdfcov Published by the American Association of Physicists in Medicine Articles you may be interested in Antiscatter grids in mobile C-arm cone-beam CT: Effect on image quality and dose Med. Phys. 39, 153 (2012); 10.1118/1.3666947 A quality assurance program for image quality of cone-beam CT guidance in radiation therapy Med. Phys. 35, 1807 (2008); 10.1118/1.2900110 Chord-based image reconstruction in cone-beam CT with a curved detector Med. Phys. 33, 3743 (2006); 10.1118/1.2337270 Cardiac cone-beam CT volume reconstruction using ART Med. Phys. 32, 851 (2005); 10.1118/1.1869052 X-ray scatter correction algorithm for cone beam CT imaging Med. Phys. 31, 1195 (2004); 10.1118/1.1711475
Improved image quality of cone beam CT scans for radiotherapy image guidance using fiber-interspaced antiscatter grid Uros Stankovic, Marcel van Herk, Lennert S. Ploeger, and Jan-Jakob Sonkea) Department of Radiation Oncology, The Netherlands Cancer Institute, Amsterdam 1066 CX, The Netherlands
(Received 7 January 2014; revised 26 March 2014; accepted for publication 27 April 2014; published 21 May 2014) Purpose: Medical linear accelerator mounted cone beam CT (CBCT) scanner provides useful soft tissue contrast for purposes of image guidance in radiotherapy. The presence of extensive scattered radiation has a negative effect on soft tissue visibility and uniformity of CBCT scans. Antiscatter grids (ASG) are used in the field of diagnostic radiography to mitigate the scatter. They usually do increase the contrast of the scan, but simultaneously increase the noise. Therefore, and considering other scatter mitigation mechanisms present in a CBCT scanner, the applicability of ASGs with aluminum interspacing for a wide range of imaging conditions has been inconclusive in previous studies. In recent years, grids using fiber interspacers have appeared, providing grids with higher scatter rejection while maintaining reasonable transmission of primary radiation. The purpose of this study was to evaluate the impact of one such grid on CBCT image quality. Methods: The grid used (Philips Medical Systems) had ratio of 21:1, frequency 36 lp/cm, and nominal selectivity of 11.9. It was mounted on the kV flat panel detector of an Elekta Synergy linear accelerator and tested in a phantom and a clinical study. Due to the flex of the linac and presence of gridline artifacts an angle dependent gain correction algorithm was devised to mitigate resulting artifacts. Scan reconstruction was performed using XVI4.5 augmented with inhouse developed image lag correction and Hounsfield unit calibration. To determine the necessary parameters for Hounsfield unit calibration and software scatter correction parameters, the Catphan 600 (The Phantom Laboratory) phantom was used. Image quality parameters were evaluated using CIRS CBCT Image Quality and Electron Density Phantom (CIRS) in two different geometries: one modeling head and neck and other pelvic region. Phantoms were acquired with and without the grid and reconstructed with and without software correction which was adapted for the different acquisition scenarios. Parameters used in the phantom study were tcup for nonuniformity and contrast-to-noise ratio (CNR) for soft tissue visibility. Clinical scans were evaluated in an observer study in which four experienced radiotherapy technologists rated soft tissue visibility and uniformity of scans with and without the grid. Results: The proposed angle dependent gain correction algorithm suppressed the visible ring artifacts. Grid had a beneficial impact on nonuniformity, contrast to noise ratio, and Hounsfield unit accuracy for both scanning geometries. The nonuniformity reduced by 90% for head sized object and 91% for pelvic-sized object. CNR improved compared to no corrections on average by a factor 2.8 for the head sized object, and 2.2 for the pelvic sized phantom. Grid outperformed software correction alone, but adding additional software correction to the grid was overall the best strategy. In the observer study, a significant improvement was found in both soft tissue visibility and nonuniformity of scans when grid is used. Conclusions: The evaluated fiber-interspaced grid improved the image quality of the CBCT system for broad range of imaging conditions. Clinical scans show significant improvement in soft tissue visibility and uniformity without the need to increase the imaging dose. © 2014 American Association of Physicists in Medicine. [http://dx.doi.org/10.1118/1.4875978] Key words: cone beam CT, image quality, scatter, image guided radiotherapy, antiscatter grids
1. INTRODUCTION The development of cone beam computed tomography (CBCT) with active-surface flat-panel imagers (FPI) has been an important step forward in development of image guided radiotherapy (IGRT).1, 2 The system provides useful soft tissue contrast for image guidance. One of the commonly encountered problems in CBCT imaging is presence of extensive scattered radiation due to wide cone angles. This leads to reduction in image quality in comparison to fan beam CT 061910-1
Med. Phys. 41 (6), June 2014
manifested as reduction in soft tissue visibility, Hounsfield unit inaccuracy and uniformity.3 In x-ray imaging, scatter-toprimary ratio (SPR) depends on the thickness of the scattering object and for CBCT imaging values of SPR exceeding 2.5 for water equivalent cylinder of 30 cm diameter are reported.4–6 Many scatter mitigation strategies have been proposed7, 8 ranging from hardware to software solutions. In clinical practice, a combination of several of these techniques is usually applied. Due to the geometry of the CBCT system, a large air-gap exists. This gap prevents photons scattered under wide
0094-2405/2014/41(6)/061910/12/$30.00
© 2014 Am. Assoc. Phys. Med.
061910-1
061910-2
Stankovic et al.: Improved CBCT imaging using fiber interspaced antiscatter grid
angles to reach the FPI. Most of the devices also use a bow-tie filter which modulates the beam so that edges of the patient are exposed to lower x-ray fluence thus reducing the generation of the scattered photons in these regions.4, 5 Finally, almost all systems have software scatter correction strategies. Combination of all these different methods was performed in a Monte Carlo study,4 which showed that software scatter subtraction can improve images when used in combination with ASG and bowtie filter, but under some conditions (high scatter to primary ratio accompanied by low primary transmission typically encountered in pelvic imaging) scatter subtraction has a negative effect on soft tissue visibility. Additionally, more advanced scatter correction algorithms have been proposed in literature,9–11 but the presence of scattered radiation still represents a problem for CBCT imaging. Antiscatter grids (ASG) are well known in the field of projection radiography as a tool for scatter reduction by absorbing scattered photons in strips made from a material with large atomic number (usually lead). Grids are described with two parameters: grid ratio and grid frequency. The grid ratio is a ratio of the height of the strips with the distance between two strips, and grid frequency is equal to the number of pairs of strips and interspacer per unit of length. While using the grid reduces the amount of scattered radiation that is detected, it also leads to reduction in the primary fluence. The transmission of scattered photons is decreased with increase of the grid ratio or grid frequency, which also decreases the primary transmission whether through increase of the grid thickness or the equivalent lead surface that is seen by the x-ray source. For this reason, and considering the other mechanisms of scatter control, specially the large air gap present,12 the question of applicability of ASG to the CBCT systems used for IGRT has remained somewhat controversial. A study by Siewerdsen et al.13 has shown that grids examined had benefits only under a certain set of conditions: large scatter to primary signal ratios and higher imaging doses. Similar results were obtained from Monte Carlo simulations,4, 6, 14 which showed that grids have slightly positive to positive effect on pelvic imaging and mildly negative to no effect in head and neck (H&N) imaging. However, new generation of fiber interspaced ASGs (Refs. 15 and 16) have allowed increasing the grid ratio while keeping transmission of primary radiation reasonably high. This has led to our investigation whether such a grid can bring more benefits for linac integrated CBCT systems compared to the current software correction strategy applied on the Elekta Synergy system. Another important aspect of this study was to examine the need for and necessary calibration steps in case examined grid would be introduced in the clinical practice.
2. MATERIALS AND METHODS 2.A. Imaging platform
Cone beam CT system integrated with a medical linear accelerator (Synergy 4.6, Elekta Oncology System, Crawley, UK, augmented with inhouse developed software) was used in this study. Due to the heavy weight of the gantry the system is Medical Physics, Vol. 41, No. 6, June 2014
061910-2
v
u
F IG . 1. ASG mounted on robotic arm of the kV FPI. On figure the coordinate system of the detector is also shown. Gridlines were aligned with u-axis.
not perfectly rigid during rotation and therefore a static point in the isocenter will not be projected at the same position on the FPI. This is known as flex and is corrected during the reconstruction. The detector has associated u0v coordinate system (Fig. 1). By offsetting panel in the direction of the u axis it is possible to extend the field of view from small (SFOV) to medium (MFOV) and large (LFOV). With SFOV the reconstructed volume is a cylinder with 25 cm height and basis diameter of 25 cm, with MFOV the reconstructed volume is a cylinder with same height but basis diameter of 40 cm. In our institute, LFOV is not used in practice. During image acquisition projection images are gained or flood-field corrected. Flood-field images are obtained by acquiring kV beam with no object in the beam.17 The system is equipped with a bowtie filter, which is always used in our clinic. The inhouse developed reconstruction software includes Hounsfield unit calibration and image lag correction.18 Gray values are calibrated using air and water values measured in an image quality phantom. A linear transform using those values is performed on the reconstructed scan. 2.B. Antiscatter grid
The tested grid (Philips Medical Systems, Eindhoven, The Netherlands) was made of lead strips with fiber interspacing, and it was enclosed in carbon fiber casing. It was custom made for the FPI used (Fig. 1). The grid ratio (ratio of the height of lead strips and the distance between two lead strips) was 21:1 and the grid frequency (the number of pairs lead+fiber per cm) was 36 l/cm. According to the specifications supplied by the manufacturer, the grid has a transmission rate for primary radiation of 70% measured at 80 kVp. Selectivity of the grid, defined as ratio of the transmission of primary and the transmission of scattered radiation was 11.9 (nominal at 80 kVp). The grid was mounted atop the FPI, replacing a default 1 mm thick aluminium cover, with grid lines oriented along the u-direction of the FPI. This orientation prevents defocusing of the grid when the robotic arm of the detector is shifted in u-direction to obtain larger field of view, Fig. 1.
061910-3
Stankovic et al.: Improved CBCT imaging using fiber interspaced antiscatter grid
2.C. Line artifacts and angle dependent flood-field calibration
As Gauntt and Barnes19 have shown, grids typically create line artifacts in projection radiography. The severity of these artifacts depends on the grid ratio, frequency, and spatial resolution of the imager. In case of perfectly rigid imaging system (i.e., without the flex), these grid lines would be removed by the flood-field correction. In the presence of the flex, however, it can be expected that the lines remain in the projection domain which leads to ring artifacts in reconstructed volumes. This effect can potentially be removed by using angle dependent flood-field correction. Therefore, the following algorithm was developed to generate angle dependent flood-field images:
r Acquire air scans with following rotation speed profile: first stationary (acquiring around 50 frames), then with medium speed (≈0.3 RPM) and slowing down to acquire at least 50 frames at the end of the arc. r Divide the whole angular range of rotation in N bins whose boundaries overlap which will give N different gain images. r Average all frames in each of these bins, and “tag” the resulting “gain” image with the angle that is equal to the middle of the bin. For gain files at 180◦ and −180◦ , those files acquired at the beginning and end of rotation are used. During image acquisition the angle-dependent gain image tagged with the angle closest to the currently acquired projection image is selected. The projection image is then divided with the selected angle-dependent gain image and multiplied with the mean of the angle-dependent gain image. To establish whether the grid under evaluation would need angle dependent flood-field correction, air scans were acquired with the rotation profile described above. The scans were then analyzed using Fourier transform as grid lines would in this case represent high frequency component of the Fourier spectrum. The relationship between the flex and the high frequency component was examined using Spearmans rank correlation coefficient ρ. 2.D. Software scatter correction
A simple scatter correction method assuming an uniform distribution of scatter on the FPI is implemented as part of the inhouse developed software. The method was originally developed for portal dosimetry and is based on estimating the amount of scatter in projection images.20 Since scatter originating from air is negligible, the first step in the software correction is to exclude the unattenuated portions of the projection images. In the second step, scatter fraction is estimated by calculating the mean signal value of the not excluded area and multiplying it with a constant. This constant, or scatter correction parameter Sc , usually takes values between 0.2 and 0.3. The resulting product is assumed to be equal to scatter contribution and is uniformly subtracted from all pixels. The values of parameter Sc used in this study are obtained by scanMedical Physics, Vol. 41, No. 6, June 2014
061910-3
ning a standard image quality phantom, such as Catphan in SFOV geometry, with an acquisition protocol that does not saturate the panel. The scan is then reconstructed with different values of the correction parameter until uniformity is obtained. The main goal of software scatter correction is to increase the uniformity of the images and improve the soft tissue visibility. The relationship between scatter correction strategy and image noise in the final reconstruction is not trivial. Apart from the scatter correction strategy, final image noise will depend on imaging dose, size of the imaged object, intrinsic SPR, applied reconstruction algorithm and filter, other necessary corrections (image lag, beam hardening, HU calibration), panel position, etc., as is shown in cascades systems analysis of CBCT imaging.21 Due to the simplicity of the scatter estimation it is possible that some pixels in the scatter corrected projections reach negative value. In this case, a non-negativity constraint is applied which ensures that all pixels have positive values by shifting the range of all pixels. This frequently occurs in case of pelvic imaging. 2.E. Phantoms and clinical cases
Quantitative evaluation of the image quality was performed using the CIRS CBCT Image Quality and Electron Density phantom (Computerized Imaging Reference Systems, Norfolk, VA), Fig. 2(a). The phantom was scanned in two different configurations: modified, “head and neck” [Fig. 2(b)], and standard, “pelvic” [Fig. 2(c)], configuration. Overview of scanning conditions is given in Table I. Air and water values used for HU calibration and Sc parameter optimization were obtained from Catphan 600 phantom (The Phantom Laboratory, Salem, MA). Scans were reconstructed with 1 mm voxels. During the test of the grid in clinical settings, the grid was used on two different machines where a large variety of patients were scanned. All scans were acquired with our standard clinical protocols with dose and image quality tailored for the corresponding image guided correction protocol. The scans were reconstructed in the clinic with software scatter correction using Sc parameters described in Table II. After successful phantom experiments, the clinical evaluation was approved following our internal procedures. Note that the clinical evaluation did not affect the imaging dose. In case of any adverse effects in the clinical imaging, the evaluation would have been stopped. 2.F. Evaluation parameters
2.F.1. Phantoms
The parameters used to evaluate image quality with the phantoms were (as defined in work by Siewerdsen et al.13 ): tcup , contrast, noise, and CNR. Cupping parameter, tcup , was calculated as tcup (%) = 100 ×
μedge − μcenter , μedge
(1)
061910-4
Stankovic et al.: Improved CBCT imaging using fiber interspaced antiscatter grid
061910-4
(a)
(b)
Acrylic S6
S1 Polystyrene
Delrin S5
LDPE
18 cm
27 cm
S2
(c)
Teflon S3
S4 Air
33 cm
F IG . 2. (a) A planning CT of CIRS phantom and marked regions of interests within the inserts. The regions S1 –S6 represent pseudo-inserts used for measuring the mean gray value of surrounding material. (b) The phantom in the modified configuration referred to as head and neck phantom. (c) The phantom in the standard configuration referred to as pelvis or pelvic phantom.
where μcenter is mean gray value in a 1×1 cm2 region of interest in the center of uniformity module of the phantom and μedge mean gray value in four ROIs of same size placed on the edge of the uniformity module. Contrast was defined as the absolute value of the difference between mean gray value in the ROI within an insert and mean gray value of the background, and noise as an average of the standard deviation within these two regions. As a surrogate for mean gray value of the surrounding six pseudo-inserts positioned between real inserts were averaged [S1 –S6 in Fig. 2(a)]. The phantom has six inserts [Fig. 2(a)], made of different materials (Table III). For CNR evaluation, polystyrene, low density polyethylene (LDPE), acrylic, and delrin inserts were used. The surrounding disk is made of plastic water. Another parameter used to evaluate different methods was CNR improvement defined as ratio of CNR when scatter correction is applied and CNR without correction. If value of this parameter is larger than one, it could be said that applied correction is beneficial. Bissonnette et al.17 have shown that it is not possible to calibrate Hounsfield units of scans in different geometries using one set of values. This is partially due to the TABLE I. Overview of scanning protocols and settings. Tube voltage was in all cases 120 kVp, arc length 360◦ . The average number of frames for the head and neck phantom was 350, and for the pelvic phantom 680. Phantom
ASG Bowtie Field of view Current [mA] Pulse [ms]
Head and Neck Head and Neck Head and Neck Head and Neck Pelvis Pelvis Pelvis Pelvis
Yes Yes No No Yes Yes No No
Yes No Yes No Yes No Yes No
Small Small Small Small Medium Medium Medium Medium
Medical Physics, Vol. 41, No. 6, June 2014
20 20 20 20 32 32 32 32
25 25 25 25 20 20 20 20
effects of the scattered radiation that depends on the imaged object and the imaging geometry. Therefore, with a better scatter correction method the linear fit between measured and nominal gray values in different geometries would be more comparable. This was investigated by looking into the parameters of linear fit, slope, and intercept, for different phantoms and different correction strategies. In addition, the quality of the fit was determined using the R2 values. 2.F.2. Clinical scans
Clinical scans were evaluated in an observational study. In this study, scans of 20 patients taken on successive days with and without the grid were compared by four radiotherapy technicians (RTT) with at least one year of experience with CBCT imaging. Patients included in the study covered a wide range of disease sites: head and neck, bladder, anal, and cervical cancer. The last three categories will be referred to in the rest of the paper as pelvic region cancers, for simplicity and because imaging region and reconstruction protocols are similar for these patients. In total, there were ten head and neck patients and ten patients with cancer in the pelvic region. These patients were selected because these are the two regions found in previous studies to show different effects of grid use: negative for head and neck and positive for pelvis. Scans of lung and breast cancer patients were not used in this study TABLE II. Values of the Sc parameter for the various combination of software correction and presence of ASG. Software Correction ASG
No Yes
No
Yes
0.0 0.0
0.24 0.025
061910-5
Stankovic et al.: Improved CBCT imaging using fiber interspaced antiscatter grid
TABLE III. Overview of inserts of CIRS phantom and their nominal HU. Insert
3. RESULTS 3.A. Angle dependent flood-field correction
Nominal HU
Air LDPE Polystyrene Acrylic Delrin Teflon
24 924 989 1144 1364 1974
because the image quality of these regions is more compromised by the breathing motion then by presence of scattered photons so effects of the scatter suppression method would be hard to determine. Observers got a pair of scans of the same patient with and without the grid and were asked to score which scan (if any) had better soft tissue visibility and uniformity in their opinion. The observers did not know which scan was acquired using the grid and which not. In case one of the two methods was scored as better, then that method would get one point, and in case of a tie both methods would get half of a point. The difference between the two methods was tested using the Wilcoxon signed rank test.
The correlation coefficient between the highest frequency component of Fourier transform of open beam projection images without angle-dependent gain calibration as function of gantry angle and the flex in u-direction was found to be ρ = 0.94, p = 0.00, Fig. 3. After applying angle dependent flood-field calibration with different numbers of images (5, 10, or 15), this correlation was removed. The value of the correlation coefficient was ρ = 0.067, ρ = 0.177, and ρ = 0.263 for N = 5, 10, or 15 images, respectively. The respective pvalues were p = 0.046, p = 0.00, and p = 0.00. Curves were normalized to one for better comparison. Application of angle dependent flood-field correction removed rings from the reconstructed volumes (Fig. 4). In this case, the correction was performed using 15 flood images. The minimal number of angle-dependent gain images needed to adequately remove most prominent rings was found to be five. The correction of less conspicuous rings required between 10 and 15 flood images. Note that in case of larger objects rings were less prominent. Throughout the study a correction using 15 images was applied.
Flex Not corrected Corrected, 5 images Corrected, 10 images Corrected, 15 images
0.0
0.2
0.4
Normalised
0.6
0.8
1.0
061910-5
−150
−100
−50
0
50
100
150
Gantry angle [deg]
F IG . 3. Amplitude of highest frequency component of FT as function of gantry angle and flex in u-direction. The no correction case corresponds to the images acquired without the angle-dependent correction. Medical Physics, Vol. 41, No. 6, June 2014
061910-6
Stankovic et al.: Improved CBCT imaging using fiber interspaced antiscatter grid
061910-6
F IG . 4. Transverse view of Catphan phantom without (a) and with (b) angle-dependent flood-field correction and (c) the difference of two images. (d)–(f) represent equivalent image for the larger CIRS phantom.
3.B. Phantom study
3.B.1. Image uniformity
The nonuniformity dropped by 90% for the head and neck phantom just by applying the grid (Table IV). This is still less than software correction alone (around 97%). The best results were obtained by combining grid and software, with 98.44% reduction in cupping. For pelvic phantom, the grid has shown best reduction of cupping with around 92% cupping reduction. Adding software both with and without the grid generated increase of signal in the center of the phantom, which represents overcorrection of the cupping effect.
The CNR of the head and neck phantom without the bowtie filter, using the combined grid and software correction had improved by on average a factor of 3 (ranging from 2.7 to 3.7). The grid alone improved CNR with a factor of 2.8 (ranging from 2.5 to 3.5) outperforming software correction, which improved the CNR on average 1.8 times (ranging from 1.7 to 2.2) the CNR [Fig. 5(d)]. [Figure 6(d)] The grid alone performed the best for the pelvic phantom with the bowtie, with an average increase of the CNR by 2.2 times (ranging from 0.92 to 4.6) followed by a factor of 2.1 for grid and software (ranging from 0.9 to 4.75). Software alone brought almost no improvement compared to no correction (CNR increased by a factor of 1.025, ranging from 0.7 to 1.3).
3.B.2. Contrast to noise ratio
The CNR of the head and neck phantom without the bowtie filter, using the combined grid and software correction strategy improved on average by a factor of 5.35 (ranging from 4.46 to 6.88 depending on the insert) [Fig. 5(c)]. With the grid alone, the improvement was on average by a factor of 2.7 (range from 2.37 to 3.33) and software alone was the least effective option with an average improvement factor of 1.74 (range from 1.49 to 1.98). For the pelvic phantom without the bowtie [Fig. 6(c)], the improvement was the highest with the combination of grid and software (2.71 times, ranging from 1.1 to 6.31) followed by grid alone with (CNR improvement averaged 2.3 times, ranging from 2.37 to 3.33). Again, the least effective option was software alone with an average improvement of 1.19 times (ranging from 1.11 to 1.21). TABLE IV. tcup for different phantoms (Bowtie present). Phantom
FOV
Head and neck Small Pelvis Medium
No correction Software Grid only Grid+software 12.9 13.7
0.3 − 6.6
Medical Physics, Vol. 41, No. 6, June 2014
1.2 1.1
0.2 − 1.83
3.B.3. CT number linearity
For the head and neck phantom reconstructions with HU calibration, all scatter correction approaches have very good correspondence between CT numbers and HU with slope of the linear fit of 1.01 for grid based correction 1.04 for the no correction case and 1.06 for the software corrected scans [Fig. 7(a)]. However, when same air and water values are used for the medium field of view scans, grid plus software has the best correspondence between measured CT numbers and expected HU values [Fig. 7(b)], with slope of the linear fit of 0.94 followed by grid with 0.86 and software alone with 0.84. 3.C. Clinical scans
In case of head and neck imaging, the difference between scans made with and without the grid correction was subtle (Fig. 8). Most of the improvements can be seen in the sagittal view, where it is possible to differentiate cervical vertebrae, and the tissue on the dorsal side has a more uniform appearance. Also, in the coronal view the tissue is more uniform in the shoulder region. In case of pelvic region imaging
061910-7
Stankovic et al.: Improved CBCT imaging using fiber interspaced antiscatter grid (a)
(b)
(c)
(d)
061910-7
F IG . 5. Contrast to noise ratio for head and neck phantom without (a) and with bowtie filter (b). CNR improvement in comparison with no correction for both cases (c) and (d). The lines in the CNR Improvement graph show the average improvement for a given scatter correction strategy. (a)
(b)
(c)
(d)
F IG . 6. Contrast to noise ratio for pelvic phantom without (a) and with bowtie filter (b). CNR improvement in comparison with no correction for both cases (c) and (d). The lines in the CNR Improvement graph show the average improvement for a given scatter correction strategy. Medical Physics, Vol. 41, No. 6, June 2014
061910-8
Stankovic et al.: Improved CBCT imaging using fiber interspaced antiscatter grid
(a)
061910-8
(b)
F IG . 7. CT number integrity and linearity for head and neck (a) and pelvic (b) phantom with bowtie filter. In the legend, linear fit parameters, slope, intercept, and R2 values, are shown. Same calibration parameters were used for both scanning geometries, showing that the grid makes HU calibration of the scans much more robust.
(Fig. 9), the most obvious improvement is the lack of artifacts related to non-negativity constraint that are most prominent in sagittal view but also in coronal and transverse view. In these cases, the noise penalty is lower than contrast improvement which leads to a better visualization of the boundary between
prostate and bladder when grid is applied. In both cases, soft tissue is better visualized in the scans obtained with the grid. The window and level were adjusted to better show the differences between two images, but were set in such a way that histograms of the images were equal.
F IG . 8. (a)–(c) Transverse, coronal, and sagittal view of head and neck cancer patient CBCT scan performed without the grid, (d)–(f) same views of the CBCT scan of same patient performed with ASG, (g)–(i) same views of the same patient on the diagnostic CT. In both cases, additional software correction with appropriate Sc parameter was performed. Window and level were chosen by manually matching the histograms of the scans. For the transverse view of the CBCT scans, three slices were averaged for better comparison with 3 mm slices of the diagnostic CT. Medical Physics, Vol. 41, No. 6, June 2014
061910-9
Stankovic et al.: Improved CBCT imaging using fiber interspaced antiscatter grid
061910-9
F IG . 9. (a)–(c) Transverse, coronal, and sagittal view of bladder cancer patient CBCT scan performed without the grid, (d)–(f) same views of the CBCT scan of same patient performed with ASG, (g)–(i) same views of the same patient on the diagnostic CT. In both cases, additional software correction with appropriate Sc parameter was performed. Window and level were chosen by manually matching the histograms of the scans. For the transverse view of the CBCT scans, three slices were averaged for better comparison with 3 mm slices of the diagnostic CT.
3.C.1. Observational study
4. DISCUSSION
The observational study has shown significant improvement when grid is used both for soft tissue visibility and for uniformity (Table V). For soft tissue visibility, in 47 cases the better scan was made with the grid, in 19 without, and in 14 cases there was no visible difference between two scans. For uniformity, the numbers are similar: in 48 cases scan with the grid was better, in 16 without, and in 16 there was no difference. When further dividing the data along the imaging regions (Table VI) there was no significant difference in case of soft tissue visibility for head and neck imaging, but there was for uniformity, and for pelvic imaging all differences were significant.
This was the first study that investigated the effect of a high performance fiber-interspaced antiscatter grid15, 16 in the context of CBCT imaging for image guided radiotherapy. It was found that image quality was improved in both pelvic and head and neck imaging. A large cohort of patients was scanned with and without the grid, which made possible to perform an observational study. Yet another important conclusion of this study is that it was possible to improve image quality without increasing imaging dose, even when using a grid with high lead content. In order to use the grid in the clinic, a simple but effective algorithm to generate and apply angle dependent flood-field correction was developed. In contrast to other studies which focused on comparison of grids alone against no correction, the focus of this study was on comparison with current clinical software scatter correction.
TABLE V. Observer study outcomes. Numbers are divided according to observers and show the number of answers for each option for each scan pair. Soft tissue visibility
Observer 1 Observer 2 Observer 3 Observer 4 Total p-value
Grid better
No grid better
16 10 13 8 47
4 2 2 11 19 0.0121
TABLE VI. Observer study outcomes by imaging regions. In this table, only totals are shown.
Uniformity
Equal
Grid better
No grid better
0 8 5 1 14
20 6 19 3 48
0 13 1 2 16
Improved image quality of cone beam CT scans for radiotherapy image guidance using fiber-interspaced antiscatter grid Uros Stankovic, Marcel van Herk, Lennert S. Ploeger, and Jan-Jakob Sonkea) Department of Radiation Oncology, The Netherlands Cancer Institute, Amsterdam 1066 CX, The Netherlands
(Received 7 January 2014; revised 26 March 2014; accepted for publication 27 April 2014; published 21 May 2014) Purpose: Medical linear accelerator mounted cone beam CT (CBCT) scanner provides useful soft tissue contrast for purposes of image guidance in radiotherapy. The presence of extensive scattered radiation has a negative effect on soft tissue visibility and uniformity of CBCT scans. Antiscatter grids (ASG) are used in the field of diagnostic radiography to mitigate the scatter. They usually do increase the contrast of the scan, but simultaneously increase the noise. Therefore, and considering other scatter mitigation mechanisms present in a CBCT scanner, the applicability of ASGs with aluminum interspacing for a wide range of imaging conditions has been inconclusive in previous studies. In recent years, grids using fiber interspacers have appeared, providing grids with higher scatter rejection while maintaining reasonable transmission of primary radiation. The purpose of this study was to evaluate the impact of one such grid on CBCT image quality. Methods: The grid used (Philips Medical Systems) had ratio of 21:1, frequency 36 lp/cm, and nominal selectivity of 11.9. It was mounted on the kV flat panel detector of an Elekta Synergy linear accelerator and tested in a phantom and a clinical study. Due to the flex of the linac and presence of gridline artifacts an angle dependent gain correction algorithm was devised to mitigate resulting artifacts. Scan reconstruction was performed using XVI4.5 augmented with inhouse developed image lag correction and Hounsfield unit calibration. To determine the necessary parameters for Hounsfield unit calibration and software scatter correction parameters, the Catphan 600 (The Phantom Laboratory) phantom was used. Image quality parameters were evaluated using CIRS CBCT Image Quality and Electron Density Phantom (CIRS) in two different geometries: one modeling head and neck and other pelvic region. Phantoms were acquired with and without the grid and reconstructed with and without software correction which was adapted for the different acquisition scenarios. Parameters used in the phantom study were tcup for nonuniformity and contrast-to-noise ratio (CNR) for soft tissue visibility. Clinical scans were evaluated in an observer study in which four experienced radiotherapy technologists rated soft tissue visibility and uniformity of scans with and without the grid. Results: The proposed angle dependent gain correction algorithm suppressed the visible ring artifacts. Grid had a beneficial impact on nonuniformity, contrast to noise ratio, and Hounsfield unit accuracy for both scanning geometries. The nonuniformity reduced by 90% for head sized object and 91% for pelvic-sized object. CNR improved compared to no corrections on average by a factor 2.8 for the head sized object, and 2.2 for the pelvic sized phantom. Grid outperformed software correction alone, but adding additional software correction to the grid was overall the best strategy. In the observer study, a significant improvement was found in both soft tissue visibility and nonuniformity of scans when grid is used. Conclusions: The evaluated fiber-interspaced grid improved the image quality of the CBCT system for broad range of imaging conditions. Clinical scans show significant improvement in soft tissue visibility and uniformity without the need to increase the imaging dose. © 2014 American Association of Physicists in Medicine. [http://dx.doi.org/10.1118/1.4875978] Key words: cone beam CT, image quality, scatter, image guided radiotherapy, antiscatter grids
1. INTRODUCTION The development of cone beam computed tomography (CBCT) with active-surface flat-panel imagers (FPI) has been an important step forward in development of image guided radiotherapy (IGRT).1, 2 The system provides useful soft tissue contrast for image guidance. One of the commonly encountered problems in CBCT imaging is presence of extensive scattered radiation due to wide cone angles. This leads to reduction in image quality in comparison to fan beam CT 061910-1
Med. Phys. 41 (6), June 2014
manifested as reduction in soft tissue visibility, Hounsfield unit inaccuracy and uniformity.3 In x-ray imaging, scatter-toprimary ratio (SPR) depends on the thickness of the scattering object and for CBCT imaging values of SPR exceeding 2.5 for water equivalent cylinder of 30 cm diameter are reported.4–6 Many scatter mitigation strategies have been proposed7, 8 ranging from hardware to software solutions. In clinical practice, a combination of several of these techniques is usually applied. Due to the geometry of the CBCT system, a large air-gap exists. This gap prevents photons scattered under wide
0094-2405/2014/41(6)/061910/12/$30.00
© 2014 Am. Assoc. Phys. Med.
061910-1
061910-2
Stankovic et al.: Improved CBCT imaging using fiber interspaced antiscatter grid
angles to reach the FPI. Most of the devices also use a bow-tie filter which modulates the beam so that edges of the patient are exposed to lower x-ray fluence thus reducing the generation of the scattered photons in these regions.4, 5 Finally, almost all systems have software scatter correction strategies. Combination of all these different methods was performed in a Monte Carlo study,4 which showed that software scatter subtraction can improve images when used in combination with ASG and bowtie filter, but under some conditions (high scatter to primary ratio accompanied by low primary transmission typically encountered in pelvic imaging) scatter subtraction has a negative effect on soft tissue visibility. Additionally, more advanced scatter correction algorithms have been proposed in literature,9–11 but the presence of scattered radiation still represents a problem for CBCT imaging. Antiscatter grids (ASG) are well known in the field of projection radiography as a tool for scatter reduction by absorbing scattered photons in strips made from a material with large atomic number (usually lead). Grids are described with two parameters: grid ratio and grid frequency. The grid ratio is a ratio of the height of the strips with the distance between two strips, and grid frequency is equal to the number of pairs of strips and interspacer per unit of length. While using the grid reduces the amount of scattered radiation that is detected, it also leads to reduction in the primary fluence. The transmission of scattered photons is decreased with increase of the grid ratio or grid frequency, which also decreases the primary transmission whether through increase of the grid thickness or the equivalent lead surface that is seen by the x-ray source. For this reason, and considering the other mechanisms of scatter control, specially the large air gap present,12 the question of applicability of ASG to the CBCT systems used for IGRT has remained somewhat controversial. A study by Siewerdsen et al.13 has shown that grids examined had benefits only under a certain set of conditions: large scatter to primary signal ratios and higher imaging doses. Similar results were obtained from Monte Carlo simulations,4, 6, 14 which showed that grids have slightly positive to positive effect on pelvic imaging and mildly negative to no effect in head and neck (H&N) imaging. However, new generation of fiber interspaced ASGs (Refs. 15 and 16) have allowed increasing the grid ratio while keeping transmission of primary radiation reasonably high. This has led to our investigation whether such a grid can bring more benefits for linac integrated CBCT systems compared to the current software correction strategy applied on the Elekta Synergy system. Another important aspect of this study was to examine the need for and necessary calibration steps in case examined grid would be introduced in the clinical practice.
2. MATERIALS AND METHODS 2.A. Imaging platform
Cone beam CT system integrated with a medical linear accelerator (Synergy 4.6, Elekta Oncology System, Crawley, UK, augmented with inhouse developed software) was used in this study. Due to the heavy weight of the gantry the system is Medical Physics, Vol. 41, No. 6, June 2014
061910-2
v
u
F IG . 1. ASG mounted on robotic arm of the kV FPI. On figure the coordinate system of the detector is also shown. Gridlines were aligned with u-axis.
not perfectly rigid during rotation and therefore a static point in the isocenter will not be projected at the same position on the FPI. This is known as flex and is corrected during the reconstruction. The detector has associated u0v coordinate system (Fig. 1). By offsetting panel in the direction of the u axis it is possible to extend the field of view from small (SFOV) to medium (MFOV) and large (LFOV). With SFOV the reconstructed volume is a cylinder with 25 cm height and basis diameter of 25 cm, with MFOV the reconstructed volume is a cylinder with same height but basis diameter of 40 cm. In our institute, LFOV is not used in practice. During image acquisition projection images are gained or flood-field corrected. Flood-field images are obtained by acquiring kV beam with no object in the beam.17 The system is equipped with a bowtie filter, which is always used in our clinic. The inhouse developed reconstruction software includes Hounsfield unit calibration and image lag correction.18 Gray values are calibrated using air and water values measured in an image quality phantom. A linear transform using those values is performed on the reconstructed scan. 2.B. Antiscatter grid
The tested grid (Philips Medical Systems, Eindhoven, The Netherlands) was made of lead strips with fiber interspacing, and it was enclosed in carbon fiber casing. It was custom made for the FPI used (Fig. 1). The grid ratio (ratio of the height of lead strips and the distance between two lead strips) was 21:1 and the grid frequency (the number of pairs lead+fiber per cm) was 36 l/cm. According to the specifications supplied by the manufacturer, the grid has a transmission rate for primary radiation of 70% measured at 80 kVp. Selectivity of the grid, defined as ratio of the transmission of primary and the transmission of scattered radiation was 11.9 (nominal at 80 kVp). The grid was mounted atop the FPI, replacing a default 1 mm thick aluminium cover, with grid lines oriented along the u-direction of the FPI. This orientation prevents defocusing of the grid when the robotic arm of the detector is shifted in u-direction to obtain larger field of view, Fig. 1.
061910-3
Stankovic et al.: Improved CBCT imaging using fiber interspaced antiscatter grid
2.C. Line artifacts and angle dependent flood-field calibration
As Gauntt and Barnes19 have shown, grids typically create line artifacts in projection radiography. The severity of these artifacts depends on the grid ratio, frequency, and spatial resolution of the imager. In case of perfectly rigid imaging system (i.e., without the flex), these grid lines would be removed by the flood-field correction. In the presence of the flex, however, it can be expected that the lines remain in the projection domain which leads to ring artifacts in reconstructed volumes. This effect can potentially be removed by using angle dependent flood-field correction. Therefore, the following algorithm was developed to generate angle dependent flood-field images:
r Acquire air scans with following rotation speed profile: first stationary (acquiring around 50 frames), then with medium speed (≈0.3 RPM) and slowing down to acquire at least 50 frames at the end of the arc. r Divide the whole angular range of rotation in N bins whose boundaries overlap which will give N different gain images. r Average all frames in each of these bins, and “tag” the resulting “gain” image with the angle that is equal to the middle of the bin. For gain files at 180◦ and −180◦ , those files acquired at the beginning and end of rotation are used. During image acquisition the angle-dependent gain image tagged with the angle closest to the currently acquired projection image is selected. The projection image is then divided with the selected angle-dependent gain image and multiplied with the mean of the angle-dependent gain image. To establish whether the grid under evaluation would need angle dependent flood-field correction, air scans were acquired with the rotation profile described above. The scans were then analyzed using Fourier transform as grid lines would in this case represent high frequency component of the Fourier spectrum. The relationship between the flex and the high frequency component was examined using Spearmans rank correlation coefficient ρ. 2.D. Software scatter correction
A simple scatter correction method assuming an uniform distribution of scatter on the FPI is implemented as part of the inhouse developed software. The method was originally developed for portal dosimetry and is based on estimating the amount of scatter in projection images.20 Since scatter originating from air is negligible, the first step in the software correction is to exclude the unattenuated portions of the projection images. In the second step, scatter fraction is estimated by calculating the mean signal value of the not excluded area and multiplying it with a constant. This constant, or scatter correction parameter Sc , usually takes values between 0.2 and 0.3. The resulting product is assumed to be equal to scatter contribution and is uniformly subtracted from all pixels. The values of parameter Sc used in this study are obtained by scanMedical Physics, Vol. 41, No. 6, June 2014
061910-3
ning a standard image quality phantom, such as Catphan in SFOV geometry, with an acquisition protocol that does not saturate the panel. The scan is then reconstructed with different values of the correction parameter until uniformity is obtained. The main goal of software scatter correction is to increase the uniformity of the images and improve the soft tissue visibility. The relationship between scatter correction strategy and image noise in the final reconstruction is not trivial. Apart from the scatter correction strategy, final image noise will depend on imaging dose, size of the imaged object, intrinsic SPR, applied reconstruction algorithm and filter, other necessary corrections (image lag, beam hardening, HU calibration), panel position, etc., as is shown in cascades systems analysis of CBCT imaging.21 Due to the simplicity of the scatter estimation it is possible that some pixels in the scatter corrected projections reach negative value. In this case, a non-negativity constraint is applied which ensures that all pixels have positive values by shifting the range of all pixels. This frequently occurs in case of pelvic imaging. 2.E. Phantoms and clinical cases
Quantitative evaluation of the image quality was performed using the CIRS CBCT Image Quality and Electron Density phantom (Computerized Imaging Reference Systems, Norfolk, VA), Fig. 2(a). The phantom was scanned in two different configurations: modified, “head and neck” [Fig. 2(b)], and standard, “pelvic” [Fig. 2(c)], configuration. Overview of scanning conditions is given in Table I. Air and water values used for HU calibration and Sc parameter optimization were obtained from Catphan 600 phantom (The Phantom Laboratory, Salem, MA). Scans were reconstructed with 1 mm voxels. During the test of the grid in clinical settings, the grid was used on two different machines where a large variety of patients were scanned. All scans were acquired with our standard clinical protocols with dose and image quality tailored for the corresponding image guided correction protocol. The scans were reconstructed in the clinic with software scatter correction using Sc parameters described in Table II. After successful phantom experiments, the clinical evaluation was approved following our internal procedures. Note that the clinical evaluation did not affect the imaging dose. In case of any adverse effects in the clinical imaging, the evaluation would have been stopped. 2.F. Evaluation parameters
2.F.1. Phantoms
The parameters used to evaluate image quality with the phantoms were (as defined in work by Siewerdsen et al.13 ): tcup , contrast, noise, and CNR. Cupping parameter, tcup , was calculated as tcup (%) = 100 ×
μedge − μcenter , μedge
(1)
061910-4
Stankovic et al.: Improved CBCT imaging using fiber interspaced antiscatter grid
061910-4
(a)
(b)
Acrylic S6
S1 Polystyrene
Delrin S5
LDPE
18 cm
27 cm
S2
(c)
Teflon S3
S4 Air
33 cm
F IG . 2. (a) A planning CT of CIRS phantom and marked regions of interests within the inserts. The regions S1 –S6 represent pseudo-inserts used for measuring the mean gray value of surrounding material. (b) The phantom in the modified configuration referred to as head and neck phantom. (c) The phantom in the standard configuration referred to as pelvis or pelvic phantom.
where μcenter is mean gray value in a 1×1 cm2 region of interest in the center of uniformity module of the phantom and μedge mean gray value in four ROIs of same size placed on the edge of the uniformity module. Contrast was defined as the absolute value of the difference between mean gray value in the ROI within an insert and mean gray value of the background, and noise as an average of the standard deviation within these two regions. As a surrogate for mean gray value of the surrounding six pseudo-inserts positioned between real inserts were averaged [S1 –S6 in Fig. 2(a)]. The phantom has six inserts [Fig. 2(a)], made of different materials (Table III). For CNR evaluation, polystyrene, low density polyethylene (LDPE), acrylic, and delrin inserts were used. The surrounding disk is made of plastic water. Another parameter used to evaluate different methods was CNR improvement defined as ratio of CNR when scatter correction is applied and CNR without correction. If value of this parameter is larger than one, it could be said that applied correction is beneficial. Bissonnette et al.17 have shown that it is not possible to calibrate Hounsfield units of scans in different geometries using one set of values. This is partially due to the TABLE I. Overview of scanning protocols and settings. Tube voltage was in all cases 120 kVp, arc length 360◦ . The average number of frames for the head and neck phantom was 350, and for the pelvic phantom 680. Phantom
ASG Bowtie Field of view Current [mA] Pulse [ms]
Head and Neck Head and Neck Head and Neck Head and Neck Pelvis Pelvis Pelvis Pelvis
Yes Yes No No Yes Yes No No
Yes No Yes No Yes No Yes No
Small Small Small Small Medium Medium Medium Medium
Medical Physics, Vol. 41, No. 6, June 2014
20 20 20 20 32 32 32 32
25 25 25 25 20 20 20 20
effects of the scattered radiation that depends on the imaged object and the imaging geometry. Therefore, with a better scatter correction method the linear fit between measured and nominal gray values in different geometries would be more comparable. This was investigated by looking into the parameters of linear fit, slope, and intercept, for different phantoms and different correction strategies. In addition, the quality of the fit was determined using the R2 values. 2.F.2. Clinical scans
Clinical scans were evaluated in an observational study. In this study, scans of 20 patients taken on successive days with and without the grid were compared by four radiotherapy technicians (RTT) with at least one year of experience with CBCT imaging. Patients included in the study covered a wide range of disease sites: head and neck, bladder, anal, and cervical cancer. The last three categories will be referred to in the rest of the paper as pelvic region cancers, for simplicity and because imaging region and reconstruction protocols are similar for these patients. In total, there were ten head and neck patients and ten patients with cancer in the pelvic region. These patients were selected because these are the two regions found in previous studies to show different effects of grid use: negative for head and neck and positive for pelvis. Scans of lung and breast cancer patients were not used in this study TABLE II. Values of the Sc parameter for the various combination of software correction and presence of ASG. Software Correction ASG
No Yes
No
Yes
0.0 0.0
0.24 0.025
061910-5
Stankovic et al.: Improved CBCT imaging using fiber interspaced antiscatter grid
TABLE III. Overview of inserts of CIRS phantom and their nominal HU. Insert
3. RESULTS 3.A. Angle dependent flood-field correction
Nominal HU
Air LDPE Polystyrene Acrylic Delrin Teflon
24 924 989 1144 1364 1974
because the image quality of these regions is more compromised by the breathing motion then by presence of scattered photons so effects of the scatter suppression method would be hard to determine. Observers got a pair of scans of the same patient with and without the grid and were asked to score which scan (if any) had better soft tissue visibility and uniformity in their opinion. The observers did not know which scan was acquired using the grid and which not. In case one of the two methods was scored as better, then that method would get one point, and in case of a tie both methods would get half of a point. The difference between the two methods was tested using the Wilcoxon signed rank test.
The correlation coefficient between the highest frequency component of Fourier transform of open beam projection images without angle-dependent gain calibration as function of gantry angle and the flex in u-direction was found to be ρ = 0.94, p = 0.00, Fig. 3. After applying angle dependent flood-field calibration with different numbers of images (5, 10, or 15), this correlation was removed. The value of the correlation coefficient was ρ = 0.067, ρ = 0.177, and ρ = 0.263 for N = 5, 10, or 15 images, respectively. The respective pvalues were p = 0.046, p = 0.00, and p = 0.00. Curves were normalized to one for better comparison. Application of angle dependent flood-field correction removed rings from the reconstructed volumes (Fig. 4). In this case, the correction was performed using 15 flood images. The minimal number of angle-dependent gain images needed to adequately remove most prominent rings was found to be five. The correction of less conspicuous rings required between 10 and 15 flood images. Note that in case of larger objects rings were less prominent. Throughout the study a correction using 15 images was applied.
Flex Not corrected Corrected, 5 images Corrected, 10 images Corrected, 15 images
0.0
0.2
0.4
Normalised
0.6
0.8
1.0
061910-5
−150
−100
−50
0
50
100
150
Gantry angle [deg]
F IG . 3. Amplitude of highest frequency component of FT as function of gantry angle and flex in u-direction. The no correction case corresponds to the images acquired without the angle-dependent correction. Medical Physics, Vol. 41, No. 6, June 2014
061910-6
Stankovic et al.: Improved CBCT imaging using fiber interspaced antiscatter grid
061910-6
F IG . 4. Transverse view of Catphan phantom without (a) and with (b) angle-dependent flood-field correction and (c) the difference of two images. (d)–(f) represent equivalent image for the larger CIRS phantom.
3.B. Phantom study
3.B.1. Image uniformity
The nonuniformity dropped by 90% for the head and neck phantom just by applying the grid (Table IV). This is still less than software correction alone (around 97%). The best results were obtained by combining grid and software, with 98.44% reduction in cupping. For pelvic phantom, the grid has shown best reduction of cupping with around 92% cupping reduction. Adding software both with and without the grid generated increase of signal in the center of the phantom, which represents overcorrection of the cupping effect.
The CNR of the head and neck phantom without the bowtie filter, using the combined grid and software correction had improved by on average a factor of 3 (ranging from 2.7 to 3.7). The grid alone improved CNR with a factor of 2.8 (ranging from 2.5 to 3.5) outperforming software correction, which improved the CNR on average 1.8 times (ranging from 1.7 to 2.2) the CNR [Fig. 5(d)]. [Figure 6(d)] The grid alone performed the best for the pelvic phantom with the bowtie, with an average increase of the CNR by 2.2 times (ranging from 0.92 to 4.6) followed by a factor of 2.1 for grid and software (ranging from 0.9 to 4.75). Software alone brought almost no improvement compared to no correction (CNR increased by a factor of 1.025, ranging from 0.7 to 1.3).
3.B.2. Contrast to noise ratio
The CNR of the head and neck phantom without the bowtie filter, using the combined grid and software correction strategy improved on average by a factor of 5.35 (ranging from 4.46 to 6.88 depending on the insert) [Fig. 5(c)]. With the grid alone, the improvement was on average by a factor of 2.7 (range from 2.37 to 3.33) and software alone was the least effective option with an average improvement factor of 1.74 (range from 1.49 to 1.98). For the pelvic phantom without the bowtie [Fig. 6(c)], the improvement was the highest with the combination of grid and software (2.71 times, ranging from 1.1 to 6.31) followed by grid alone with (CNR improvement averaged 2.3 times, ranging from 2.37 to 3.33). Again, the least effective option was software alone with an average improvement of 1.19 times (ranging from 1.11 to 1.21). TABLE IV. tcup for different phantoms (Bowtie present). Phantom
FOV
Head and neck Small Pelvis Medium
No correction Software Grid only Grid+software 12.9 13.7
0.3 − 6.6
Medical Physics, Vol. 41, No. 6, June 2014
1.2 1.1
0.2 − 1.83
3.B.3. CT number linearity
For the head and neck phantom reconstructions with HU calibration, all scatter correction approaches have very good correspondence between CT numbers and HU with slope of the linear fit of 1.01 for grid based correction 1.04 for the no correction case and 1.06 for the software corrected scans [Fig. 7(a)]. However, when same air and water values are used for the medium field of view scans, grid plus software has the best correspondence between measured CT numbers and expected HU values [Fig. 7(b)], with slope of the linear fit of 0.94 followed by grid with 0.86 and software alone with 0.84. 3.C. Clinical scans
In case of head and neck imaging, the difference between scans made with and without the grid correction was subtle (Fig. 8). Most of the improvements can be seen in the sagittal view, where it is possible to differentiate cervical vertebrae, and the tissue on the dorsal side has a more uniform appearance. Also, in the coronal view the tissue is more uniform in the shoulder region. In case of pelvic region imaging
061910-7
Stankovic et al.: Improved CBCT imaging using fiber interspaced antiscatter grid (a)
(b)
(c)
(d)
061910-7
F IG . 5. Contrast to noise ratio for head and neck phantom without (a) and with bowtie filter (b). CNR improvement in comparison with no correction for both cases (c) and (d). The lines in the CNR Improvement graph show the average improvement for a given scatter correction strategy. (a)
(b)
(c)
(d)
F IG . 6. Contrast to noise ratio for pelvic phantom without (a) and with bowtie filter (b). CNR improvement in comparison with no correction for both cases (c) and (d). The lines in the CNR Improvement graph show the average improvement for a given scatter correction strategy. Medical Physics, Vol. 41, No. 6, June 2014
061910-8
Stankovic et al.: Improved CBCT imaging using fiber interspaced antiscatter grid
(a)
061910-8
(b)
F IG . 7. CT number integrity and linearity for head and neck (a) and pelvic (b) phantom with bowtie filter. In the legend, linear fit parameters, slope, intercept, and R2 values, are shown. Same calibration parameters were used for both scanning geometries, showing that the grid makes HU calibration of the scans much more robust.
(Fig. 9), the most obvious improvement is the lack of artifacts related to non-negativity constraint that are most prominent in sagittal view but also in coronal and transverse view. In these cases, the noise penalty is lower than contrast improvement which leads to a better visualization of the boundary between
prostate and bladder when grid is applied. In both cases, soft tissue is better visualized in the scans obtained with the grid. The window and level were adjusted to better show the differences between two images, but were set in such a way that histograms of the images were equal.
F IG . 8. (a)–(c) Transverse, coronal, and sagittal view of head and neck cancer patient CBCT scan performed without the grid, (d)–(f) same views of the CBCT scan of same patient performed with ASG, (g)–(i) same views of the same patient on the diagnostic CT. In both cases, additional software correction with appropriate Sc parameter was performed. Window and level were chosen by manually matching the histograms of the scans. For the transverse view of the CBCT scans, three slices were averaged for better comparison with 3 mm slices of the diagnostic CT. Medical Physics, Vol. 41, No. 6, June 2014
061910-9
Stankovic et al.: Improved CBCT imaging using fiber interspaced antiscatter grid
061910-9
F IG . 9. (a)–(c) Transverse, coronal, and sagittal view of bladder cancer patient CBCT scan performed without the grid, (d)–(f) same views of the CBCT scan of same patient performed with ASG, (g)–(i) same views of the same patient on the diagnostic CT. In both cases, additional software correction with appropriate Sc parameter was performed. Window and level were chosen by manually matching the histograms of the scans. For the transverse view of the CBCT scans, three slices were averaged for better comparison with 3 mm slices of the diagnostic CT.
3.C.1. Observational study
4. DISCUSSION
The observational study has shown significant improvement when grid is used both for soft tissue visibility and for uniformity (Table V). For soft tissue visibility, in 47 cases the better scan was made with the grid, in 19 without, and in 14 cases there was no visible difference between two scans. For uniformity, the numbers are similar: in 48 cases scan with the grid was better, in 16 without, and in 16 there was no difference. When further dividing the data along the imaging regions (Table VI) there was no significant difference in case of soft tissue visibility for head and neck imaging, but there was for uniformity, and for pelvic imaging all differences were significant.
This was the first study that investigated the effect of a high performance fiber-interspaced antiscatter grid15, 16 in the context of CBCT imaging for image guided radiotherapy. It was found that image quality was improved in both pelvic and head and neck imaging. A large cohort of patients was scanned with and without the grid, which made possible to perform an observational study. Yet another important conclusion of this study is that it was possible to improve image quality without increasing imaging dose, even when using a grid with high lead content. In order to use the grid in the clinic, a simple but effective algorithm to generate and apply angle dependent flood-field correction was developed. In contrast to other studies which focused on comparison of grids alone against no correction, the focus of this study was on comparison with current clinical software scatter correction.
TABLE V. Observer study outcomes. Numbers are divided according to observers and show the number of answers for each option for each scan pair. Soft tissue visibility
Observer 1 Observer 2 Observer 3 Observer 4 Total p-value
Grid better
No grid better
16 10 13 8 47
4 2 2 11 19 0.0121
TABLE VI. Observer study outcomes by imaging regions. In this table, only totals are shown.
Uniformity
Equal
Grid better
No grid better
0 8 5 1 14
20 6 19 3 48
0 13 1 2 16
