Cone-beam computed tomography (CBCT) images have inaccurate CT numbers because of scattered photons. Thus, quantitative analysis of scattered photons that affect an electron density (ED) curve and calculated doses may be effective information to achieve CBCT-based radiation treatment planning. We quantitatively evaluated the effect of scattered photons on the accuracy of dose calculations from a lung image. The Monte Carlo method was used to calculate CBCT projection data, and we made two calibration curves for conditions with or without scattered photons. Moreover, we applied cupping artifact correction and evaluated the effects on image uniformity and dose calculation accuracy. Dose deviations were compared with those of convention al CT in conventional and volumetric intensity modulated arc therapy (VMAT) planning by using γ analysis and dose volume histogram (DVH) analysis. We found that cupping artifacts contaminated the scattered photons, and the γ analysis showed that the dose distribution was most decreased for a scattered photon ratio of 40%. Cupping artifact correction significantly improved image uniformity; therefore, ED curves were near ideal, and the pass rate results were significantly higher than those associated with the scattered photon effect in 65.1% and 78.4% without correction, 99.5% and 97.7% with correction, in conventional and VMAT planning, respectively. In the DVH analysis, all organ dose indexes were reduced in the scattered photon images, but dose index error rates with cupping artifact correction were improved within approximately 10%. CBCT image quality was strongly affected by scattered photons, and the dose calculation accuracy based on the CBCT image was improved by removing cupping artifacts caused by the scattered photons .
A kilo-voltage cone-beam computed tomography (CBCT) system mounted on a linear accelerator has become available for image guided radiotherapy. The main purpose of CBCT is to improve the accuracy of target localization in radiation therapy [
In previous studies, degradation of accuracy in calculated dose distribution has been shown when original CBCT numbers were used for dose calculations; therefore, replacement of CBCT numbers with conventional CT numbers has been suggested [
If we could reconstruct a CBCT image by using accurate Hounsfield unit values, we could improve the accuracy of dose calculations by using a quantitative image value without having to use several CBCT number replacement methods. Therefore, this approach can save time in the CBCT based re-planning and adaptive planning process. However, image quality of CBCT was affected by the scattered photon and the beam hardening artifact. In previous studies on scattered photon correction methods for CBCT images, several algorithmic methods have been proposed for estimating and correcting scattered photons in the CBCT projection data. Zhu et al. proposed the blocker-based method that allows approximate characterization of the scattered photons in the shadow of the blocker [
Adequate level of scattered photon reduction in the CBCT image for accurate dose calculation can be a goal of these CBCT image improvement methods, however, knowledges of amount for reduction of scattered radiation to achieve dose calculation with acceptable precision accuracy using CBCT has not been known. Thus, quantitative analysis of the scattered photons that affect a calibration curve of CBCT number to electron densities and calculated doses may be effective information for achieving CBCT-based radiation treatment planning. This type of investigation is possible to reveal the CBCT image quality for accurate dose calculations in patients.
In this study, we quantitatively evaluated the effect of scattered photons on the accuracy of dose calculations by using CBCT images. We simulated the CBCT projection data of an electron density (ED) phantom by using the MC method and made calibration curves of CBCT pixel numbers to several electron densities, including various ratios of the scattered photons. Similarly, we calculated patient lung images and converted CBCT pixel numbers to electron densities by using these calibration curves. Additionally, we determined the relationship between the accuracy of calculated patient dose and the amount of the scattered photons in CBCT images. Moreover, the effects of cupping artifact correction on CBCT images and on image uniformity were also investigated along with the dose calculation accuracy for CBCT images with cupping artifact correction.
The MC method was used to calculate CBCT projection data of an ED phantom. This calculation system was in-house program, and we calculated transport of only photons with kilovolt energy using a random number generator with the Mersenne twister for calculating probability of interactions [
Tube voltage | 100 kV |
---|---|
Number of photons | 100,000 |
Data acquisition angle | 360˚ |
Number of projections | 180 |
Detector matrix | 400 × 300 pixels |
Pixel size | 1.0 × 1.0 mm2 |
Calculation grid | 1.0 mm |
Cutoff energy | 30 keV |
Random generator | Mersenne twister |
Recon algorithm | FBP |
Recon filter | Shepp & Logan |
x-ray tube with a voltage of 100 kV, and we acquired 180 projection images of 2˚ each over 360˚. Regarding the interactions of photons with materials, we considered compton scattering, coherent scattering, and the photoelectric effect, and we defined the scattered photons as the photons involved in these interactions. The calculation accuracy of this MC simulation confirmed that counts of attenuated photons and energy spectrum coincide with the theoretically calculated value with an accuracy within 1.0%.
For the CBCT pixel number conversion curves of the ED phantom, which consisted of the ten density specification materials (air, lung, adipose, breast, water, soft tissue, muscle, liver, mineral bone, and cortical bone) were reconstructed with the MC projection data. The diameter of the ED phantom was 22 cm, and each material was 2.0 cm. The composition of the ED phantom and the details of each material are shown in
We then made two types of calibration curves of CBCT pixel values to several electron densities: one was a condition that included various ratios of scattered photons and another one without scattered photons, which curve means equivalent one acquired by the multi-slice CT (MSCT). Because, detector is strictly collimated in MSCT system, thus scattered photon contaminations are very small. The ratios of the scattered photon contaminations for each curve were successively reduced by 20% in 100% - 20% from the ratio used in the previous calibra-
Mass density (g/cm3) | Electron density | |
---|---|---|
Air | 0.00 | 0.00 |
Lung | 0.28 | 0.29 |
Adipose | 0.94 | 0.93 |
Breast | 0.98 | 0.96 |
Water | 1.00 | 1.00 |
Soft tissue | 1.02 | 1.01 |
Muscle | 1.05 | 1.02 |
Liver | 1.10 | 1.06 |
Mineral bone | 1.15 | 1.10 |
Cortical bone | 1.82 | 1.69 |
tion curve, moreover 10% and 5% - 1% scattered photon contaminations curves were obtained. Additionally, we set ten circular regions of interest (ROIs), and obtained the mean pixel value in the reconstructed image. The size of the circular ROI was 10 pixels. Using these conversion curves, we associated each ED value with the corresponding CBCT pixel value.
To compare the effect of image quality on the accuracy of the calculated doses from the CBCT image, we acquired projection data of the lung images treated by radiation therapy for lung cancer by using the MC method under the same conditions in section 2.1, and reconstructed the lung images under each condition by including the scattered photons corresponding to each calibration curve. In this simulation, the MSCT of the lung image, which was acquired in previously, was divided into three organs (lungs, bones, and soft tissues), and linear attenuation coefficients were obtained from these three areas. Projection data of this lung image attenuated by each region according to the linear attenuation coefficients were calculated and counted with the MC method.
For these lung image studies, dose distributions were calculated from CBCT images after conversion of the CBCT pixel values by using the calibration curve. As a reference dose distribution, the MSCT planning was calculated by using the standard CT value calibration curve acquired from the MSCT system.
Beam hardening and scattered photons each reduced the measured attenuation coefficients. As a result, beam hardening and scattered photons produced a common artifact known as the cupping artifact [
To correct the cupping artifact on the CBCT reconstructed image, a simple correction method assuming a uniform water object was used. In this method, we calculated ideal projection data (Rj) with an effective spectrum (Em) from the transmission distance (L(j)) of the water object in each projection, as shown in Equation (2). In cupping artifact correction, pixel values with the cupping artifact were modified according to these ideal projected dates, therefore, both the effect of scattered photons and beam hardening were corrected with this method. Then, we created cupping artifact correction maps in every ratio of the scattered photons, and the ED phantom image and patient lung image with cupping artifact were modified using these correction maps in each scattered photon classes.
I: number of photons
µw: attenuation coefficient with Em
Moreover, the ED phantom image and the lung image were corrected for the cupping artifact, and the accuracies of the calculated doses were evaluated. In correction of the lung image, the inside of the lung image contour was replaced with water material, and similarly, correction maps in every ratio of the scattered photon contaminations were created, and the cupping artifact in lung images were modified.
The differences between the dose distributions calculated from the lung CBCT image with scattered photons and from the image without scattered photons were analyzed by using the γ analysis. The criteria for the pass rate in the γ analysis were 3% of the absolute dose and ≤3 mm agreement. In this evaluation, five- field beam conventional planning and 1-arc volumetric arc therapy (VMAT) planning (coplanar, 6-MV x-rays) were calculated by using the Eclipse ver. 13.6 treatment planning system (Varian Medical Systems, USA), and the anisotropic analytical algorithm (AAA) and the Acuros XB algorithm were used for the dose calculation algorithm. The prescribed dose was 200 cGy for 95% of the target volume, and each structural contour was common size and location in each plan. Moreover, the dose distributions were evaluated by using dose volume histogram (DVH) analysis, and dose differences among the scattered photons in this lung image were expressed as the Dmean and D95 in each target area and the V20, V5, D1cc, and Dmean in several organs at risk (OARs). In these evaluations, the differences between the dose distributions based on the MSCT image and those based on the CBCT images with scattered photons were analyzed. Therefore, the contour of the target organs and OARs, planning geometry, and number of monitor units remained unchanged.
these curves approached the primary photon curve; thus, the differences between the ED values corresponding to the pixel values were reduced.
Primary photons | with scattered photons | |||||||
---|---|---|---|---|---|---|---|---|
5% | 10% | 20% | 40% | 60% | 80% | 100% | ||
tcup [%] | 0.8 | 23.7 | 30.1 | 34.1 | 37.7 | 35.0 | 34.0 | 33.2 |
the dose distribution in the MSCT planning. The pass rate of the primary photon image was near 100%; on the other hand, the pass rates of the images with scattered photons were significantly smaller than the pass rate of the primary photon image. Moreover, the pass rates of the images with cupping artifact correction were improved and did not depend on the amount of scattered photons.
A CBCT imaging device attached to a linear accelerator can be used for correct-
Parameter | Error rate [%] | ||||||||
---|---|---|---|---|---|---|---|---|---|
primary | scatter 5% | scatter 10% | scatter 20% | scatter 40% | scatter 60% | scatter 80% | scatter 100% | ||
Conventional | PTV D95 (%) | 2.0 | −7.8 | −11.8 | −14.5 | −15.2 | −12.9 | −12.5 | −11.7 |
PTV Dmean (%) | 0.0 | −8.1 | −11.9 | −14.4 | −15.2 | −13.0 | −12.3 | −11.5 | |
CTV D95 (%) | 1.5 | −8.4 | −12.5 | −15.0 | −15.9 | −13.7 | −13.4 | −12.9 | |
CTV Dmean (%) | −0.8 | −8.1 | −11.9 | −14.5 | −15.3 | −13.2 | −12.4 | −11.4 | |
Lung V20 (Gy) | −5.3 | −18.1 | −24.3 | −28.2 | −29.3 | −25.6 | −24.5 | −23.3 | |
Lung V5 (Gy) | −6.8 | −10.1 | −12.1 | −13.3 | −13.5 | −12.1 | −11.3 | −10.6 | |
Cord D1cc (Gy) | −4.3 | −11.3 | −15.1 | −18.0 | −19.4 | −16.4 | −14.8 | −12.4 | |
Cord Dmean (%) | −3.8 | −9.2 | −13.0 | −15.7 | −17.3 | −14.6 | −13.0 | −10.8 | |
VMAT | PTV D95 (%) | 3.8 | −6.1 | −8.9 | −9.8 | −10.2 | −9.8 | −11.4 | −17.4 |
PTV Dmean (%) | −0.5 | −7.6 | −10.5 | −11.7 | −11.9 | −11.0 | −10.9 | −12.7 | |
CTV D95 (%) | 0.6 | −7.7 | −10.5 | −11.9 | −12.7 | −11.3 | −11.5 | −14.2 | |
CTV Dmean (%) | −1.7 | −8.5 | −11.5 | −13.0 | −13.3 | −12.3 | −11.7 | −12.5 | |
Lung V20 (Gy) | −4.2 | −15.5 | −19.7 | −21.4 | −21.4 | −20.6 | −21.1 | −25.9 | |
Lung V5 (Gy) | −4.6 | −10.6 | −13.5 | −15.3 | −15.6 | −14.9 | −14.6 | −16.1 | |
Cord D1cc (Gy) | −4.1 | −11.4 | −14.9 | −16.9 | −16.7 | −15.3 | −13.7 | −13.5 | |
Cord Dmean(%) | −2.5 | −10.0 | −13.4 | −14.9 | −14.9 | −13.4 | −11.9 | −11.4 |
Parameter | Error rate [%] | |||||||
---|---|---|---|---|---|---|---|---|
scatter 5% | scatter 10% | scatter 20% | scatter 40% | scatter 60% | scatter 80% | scatter 100% | ||
Conventional | PTV D95 (%) | −3.3 | −4.0 | −3.3 | −1.8 | −1.6 | −2.4 | −4.4 |
PTV Dmean (%) | −3.6 | −4.0 | −3.1 | −1.7 | −1.6 | −2.2 | −3.6 | |
CTV D95 (%) | −3.7 | −4.5 | −4.0 | −2.6 | −2.4 | −3.3 | −5.2 | |
CTV Dmean (%) | −3.7 | −4.1 | −3.3 | −2.1 | −2.0 | −2.4 | −3.4 | |
Lung V20 (Gy) | −11.6 | −11.8 | −10.9 | −8.6 | −8.5 | −9.9 | −12.0 | |
Lung V5 (Gy) | −8.7 | −9.5 | −9.3 | −8.7 | −8.8 | −9.2 | −9.3 | |
Cord D1cc (Gy) | −7.6 | −8.7 | −8.5 | −7.7 | −7.3 | −7.4 | −7.5 | |
Cord Dmean (%) | −5.9 | −7.0 | −7.0 | −6.5 | −5.9 | −6.5 | −5.9 | |
VMAT | PTV D95 (%) | −2.3 | −3.0 | −2.4 | −1.0 | −1.3 | −3.2 | −9.0 |
PTV Dmean (%) | −4.0 | −4.5 | −4.2 | −3.4 | −3.4 | −4.3 | −6.2 | |
CTV D95 (%) | −4.0 | −5.2 | −5.1 | −3.9 | −4.2 | −5.7 | −9.2 | |
CTV Dmean (%) | −5.0 | −6.0 | −5.9 | −5.3 | −5.3 | −6.1 | −7.4 | |
Lung V20 (Gy) | −10.0 | −11.3 | −10.0 | −8.7 | −9.2 | −10.3 | −14.2 | |
Lung V5 (Gy) | −7.9 | −10.8 | −8.9 | −7.8 | −10.4 | −10.5 | −10.1 | |
Cord D1cc (Gy) | −7.8 | −8.9 | −9.2 | −8.6 | −8.5 | −8.8 | −9.3 | |
Cord Dmean (%) | −6.5 | −7.5 | −8.0 | −7.0 | −7.0 | −7.5 | −8.0 |
ing the patient set-up position, verifying a target position, and observing deformation of a tumor and normal tissues. However, a CBCT image is commonly degraded by scattered photons and thus, cannot be directly used for dose calculation. In this study, to evaluate the effect of the scattered photons for achieving CBCT-based dose calculation, we investigated the effect of scattered photons on calibration curves of CT values to electron densities and calculated doses by using the MC method.
For the calculated dose distribution results in the lung image, the dose distribution in the primary photon image (
The CBCT images were strongly affected by scattered photons, and the calibration table of electron density was inaccurate. In the conventional planning, the relative electron density values from the pixel values in the CBCT image was used for the AAA dose calculation algorithm. In the VMAT planning, the mass density values from the pixel values in the CBCT images were used for the Acuros XB dose calculation algorithm. In both calculation algorithms, the calculation accuracies using the CBCT images with beam-hardening artifacts were degraded; therefore, the quality of the CBCT images needed to be corrected to achieve CBCT-based dose calculation. To achieve highly accurate CBCT dose calculations, scattered photons contamination should be maintain by approximately <5% of the total variation. Therefore, applying the scattered photon correction method and development of the anti-scatter grid for the CBCT system, which can block the scattered photons contaminating the detector, was useful for achieving CBCT-based ART. Moreover, because cupping artifact correction significantly improved dose calculation accuracy, this method was found to be an effective correction technique for dose calculations using CBCT images.
The influences of the fluctuation of MC calculation data were small for results of dose calculation with scattered photon contamination, since the accuracy of MC simulation within 1.0%. In this study, objectives (ED phantom and lung images) assumed the use of uniform water material, and the ideal pixel value with a mono-beam spectrum and beam-hardening correction was calculated. If we assume the use of an object with several organ regions and calculate the ideal pixel value for each organ, the accuracy of the ED curve and dose distribution can be improved. However, accurate segmentation of each organ in the clinical CBCT images is difficult, because, the CBCT image have large fractionations by the scattered photon and motion artifact. And, inaccurate segmentation leads to cause wrong passing projection lengths and pixel values. Therefore, accurate segmentation method for improving the cupping artifact correction method in the CBCT image is needed for improving the accuracy of the ED curve and dose calculation.
We quantitatively analyzed the effect of scattered photons on dose calculations. The quality of CBCT images was strongly affected by scattered photons and should be corrected to achieve CBCT-based dose calculation. We also demonstrated that the accuracy of the CBCT-based dose calculation can be improved by removing scattered photons in approximately <5% of the total variation and applying the cupping artifact correction method.
Usui, K., Ogawa, K. and Sasai, K. (2017) Analysis of Dose Cal- culation Accuracy in Cone Beam Computed Tomography with Various Amount of Sca- ttered Photon Contamination. International Journal of Medical Physics, Clinical Engi- neering and Radiation Oncology, 6, 233-251. https://doi.org/10.4236/ijmpcero.2017.63022