An additional tilted‐scan‐based CT metal‐artifact‐reduction method for radiation therapy planning

Abstract Purpose As computed tomography (CT) imaging is the most commonly used modality for treatment planning in radiation therapy, metal artifacts in the planning CT images may complicate the target delineation and reduce the dose calculation accuracy. Although current CT scanners do provide certain correction steps, it is a common understanding that there is not a universal solution yet to the metal artifact reduction (MAR) in general. Particularly noting the importance of MAR for radiation treatment planning, we propose a novel MAR method in this work that recruits an additional tilted CT scan and synthesizes nearly metal‐artifact‐free CT images. Methods The proposed method is based on the facts that the most pronounced metal artifacts in CT images show up along the x‐ray beam direction traversing multiple metallic objects and that a tilted CT scan can provide complementary information free of such metal artifacts in the earlier scan. Although the tilted CT scan would contain its own metal artifacts in the images, the artifacts may manifest in a different fashion leaving a chance to concatenate the two CT images with the metal artifacts much suppressed. We developed an image processing technique that uses the structural similarity (SSIM) for suppressing the metal artifacts. On top of the additional scan, we proposed to use an existing MAR method for each scan if necessary to further suppress the metal artifacts. Results The proposed method was validated by a simulation study using the pelvic region of an XCAT numerical phantom and also by an experimental study using the head part of the Rando phantom. The proposed method was found to effectively reduce the metal artifacts. Quantitative analyses revealed that the proposed method reduced the mean absolute percentages of the error by up to 86% and 89% in the simulation and experimental studies, respectively. Conclusions It was confirmed that the proposed method, using complementary information acquired from an additional tilted CT scan, can provide nearly metal‐artifact‐free images for the treatment planning.

stage prostate cancer patient with bilateral hip implants. 4 The IMRT plans involved a tedious task of preventing the beams from passing through the metals. However, IMRT plans still delivered more than 105% of the prescribed dose to the target volume. A homogenous pelvic phantom-based study by Ding and Yu showed an underestimation of CT numbers of metal implants. 5 Without any corrections to the CT image, an overestimated dose was calculated by a commercial analytical 3D treatment planning system (CADPLAN) and was delivered to the target volume. Inaccuracy of CT numbers is one of the common problems encountered by medical physicists in constructing successful treatment plans for patients. For instance, dosimetric errors from 6 and 18 MV radiotherapy (RT) plans in the target volume of prostate patients with bilateral hip prostheses were reported by Wei et al. 6 The target volume for both photon beam energies was undermined due to the metal artifact-corrupted CT images used in the treatment planning. Also, 6 MV four-field RT plans were shown to be more susceptible to the metal artifacts than 18 MV four-field RT plans. Likewise, OARs are considered as major concerns in RT planning. Without metal artifact suppression, the target volume and OARs receive underestimated dose and overestimated dose, respectively. These dose perturbations were also manifested in the Monte Carlo (MC) dose calculations of bilateral prostheses phantom studies and prostate patient study at the same photon energies conducted by Bazalova et al. 7 After implementing a sinogram inpainting-based MAR, identification and delineation of the target tumor and OARs became more straightforward than utilizing the artifact-contaminated CT images. With an additional extended calibration to aid and increase the MC dose calculation accuracy, the dosimetric error in 6 MV RT plan dropped from 25% in uncorrected CT images to about 2% in MAR-corrected CT images. The improvement in dose calculations was also seen in the 18 MV case. These dose perturbation issues arising from the pelvic irradiation of patients with hip prostheses were also addressed in Task Group 63 of Association of Physicists in Medicine (AAPM) Radiation Therapy Committee (RTC). 8 Generally in clusters and small in size, highly attenuating dental filling materials (DFMs) considerably influence the CT images of oral cavity and head-and-neck (H&N) regions. Calculated dose from CT images of these regions resulted in significant dose increase to the OARs due to backscatter from the DFMs and decrease in the target tumor coverage due to the high attenuation property of DFMs.
These anomalies were reduced after applying a mask that forces the metal streak artifacts to a soft tissue value of 10 HU, and applying a virtual filter that compensates for the beam attenuation of DFMs.
Specifically, the mask improved the dose homogeneity while the virtual filter enhanced the delivered dose to the target tumor. These findings were obtained from the phantom and patient studies using RapidArc RT plans in the successive studies by Mail et al. 9,10 In another dental phantom study by Maerz et al., dose distribution deviations were calculated in both IMRT and volumetric modulated arc therapy (VMAT) plans generated from metal artifact-contaminated CT images. 11 Their study also concluded that RT plans created from metal artifact-corrected CT images resulted in a significant decrease in dose perturbations for the H&N cases. In terms of accuracy of dose calculations, VMAT exhibited a closer dose distribution agreement with the reference film measurement data than IMRT.
Spine implants are low-or high-Z metals with a complex geometry, usually situated within or near the target volume. Therefore, delineation of both target volume and OARs has always been a difficult task due to the metal artifacts. Son et al. indicated that an average of 2% dose calculation discrepancy between the implants and increasing dose errors toward the location of an implant were observed in their phantom study. 12 A clinical study conducted by Spadea et al. revealed that low-and high-Z metal implants affect dose perturbations differently in uncorrected CT images. 13 An MAR approach, incorporating the metal material information, developed by Verburg and Seco was implemented to reduce metal artifacts in the patient CT images. 14 For low-Z metal spine implants, no significant dose discrepancy was exhibited between IMRT plans created from artifact-contaminated and artifact-corrected CT images. For high-Z metal implant, however, that is, gold dental fillings and platinum wire for artery embolization, dose errors as high as 20-25% were calculated near the implants.
Compared to the x-ray external beam radiation therapy (EBRT) and brachytherapy, treatment plans for proton therapy and heavy ion therapy substantially rely on accurate stopping power derived from CT numbers of materials along the beam path to calculate the dose distribution and beam ranges. However, CT number accuracy decreases with the presence of metal artifacts inducing errors in the target coverage and lessening the sparing of normal tissues. Phantom studies by Jakel and Reiss exhibited that the metal artifacts alone generated by dental fillings, titanium hip implants, and steel hip implants underestimated the ion beam range by as much as 3%, up to 5% and 18%, respectively. 15 Verburg and Seco reported that errors in the beam range caused by titanium spine implants were also dictated by the geometry of the implant and proton beam orientation relative to the implant and artifacts. 16 In their phantom study, because none is able to completely remove metal artifacts in every situation. 41 Although existing methods may remove metal artifacts in some cases, they may introduce new artifacts or false structures, or even degrade image quality, in other cases.
This study proposes a novel approach, which utilizes data from an additional tilted CT scan, to MAR. This new method is based on the fact that most metal artifacts in CT images are caused by the object's high attenuation on traversing beams in CT scans. Therefore, tilted CT scans would provide information complementary to that of scans in which some regions are free of metal artifacts. Using the two images, a combined artifact-free image with much reduced metal artifacts can be generated. This study utilized a modified version of structural similarity (SSIM) as an index to select the regions with less metal artifacts. Quantitative analyses in both simulations and experiments were conducted to show that the proposed method effectively reduces metal artifacts in the reconstructed images.

2.A | Idea
The main idea underlying the proposed method is that tilted CT scans can provide complementary image data free of metal artifacts in the regions that have been contaminated in the original CT image.
Metal artifacts appear different in the reconstructed images obtained from CT scans at varying system-tilt angles. 45 These differences are due to the effects of physical factors that cause metal artifacts, including photon starvation, beam hardening, scatter, and noise, all of which are subject to change as the scanner tilt angle is altered. Therefore, tilted CT scans can provide information complementary to that of the standard CT scans, with the tilted CT images being free of metal artifacts and the standard CT images containing the metal artifacts. By selecting regions with less metal artifacts between the two images, images nearly free of artifacts can be generated by combining the two CT images.
A reconstructed image at an ordinary 0-degree gantry tilt-angle contains the metal artifacts of an ordinary scan, whereas an image acquired at a tilted angle may be composed of the artifact-free image in the contaminated regions in an ordinary scan and the metal artifacts from an oblique scan. Because the object structures in the two images would be nearly identical, differences between the two images would be due only to the metal artifacts. Therefore, difference between the two reconstructed images would constitute a superposition map of metal artifacts from a standard CT and an oblique CT scan. This superposition map would have no structural information about the scanned object, but would only contain the superposition of metal artifacts from the two images. 46 The correlation maps from each reconstructed image and superposition map would describe the degree of contained artifacts in the relevant reconstructed image. That is, a higher value in the correlation map KIM ET AL.
| 239 would indicate that the corresponding reconstructed image contains more artifacts. Therefore, the regions chosen from the two correlation maps with lower correlation values would be a template for artifact-free image.
A modified version of SSIM was used as an index to calculate the degree of correlation between the reconstructed images and their corresponding artifact superposition maps. 47 SSIM was designed to calculate the similarity between two images by measuring three types of visual perception: luminance, contrast, and structure.
SSIMðx; yÞ ¼ lðx; yÞ α cðx; yÞ β sðx; yÞ γ ; (1) where lðx; yÞ α , cðx; yÞ β , and sðx; yÞ γ are luminance, contrast, and structure factors, respectively. The individual factors can be calculated as: where μ y , σ 2 , and σ xy are the average, variance, and covariance, respectively. The luminance factor is associated with the average value or intensity of each image; the contrast factor is associated with the variance of each image; and the structure factor is associ-

2.B.2 | Artifact splitting
After denoising, an artifact superposition map is synthesized to exclude structural information on the scanned object; only metal artifacts were considered, as mentioned in Section 2.A. This artifact superposition map was constructed by calculating the difference between the two CT images. As shown in Fig. 1(c), the difference between the original and the tilted CT images represents the superposition of metal artifacts from the two images.

2.B.4 | Generation of artifact-reduced images
As higher values in the correlation map represent greater contamination with metal artifacts in the corresponding reconstructed images, selecting the regions with lower correlation values would lead to relatively more artifact-free than the two CT images. Therefore, the final artifact-free image can be generated using Eq. (6):

2.C | Experimental conditions
Both simulation and experimental studies were performed to determine the feasibility of the proposed method and to compare its performance with those of existing methods.

2.C.1 | Simulation study
The simulation study was performed using the pelvic region of an XCAT numerical phantom. 48  inserted into the phantom. Projection data were acquired at gantry tilt angles equal to 0°and 10°to avoid overlapping of metal implants along the beam direction (Fig. 2). The nature of polychromatic x rays was simulated by summing the weighted data of six monochromatic x rays at representative energy bins as shown in Fig. 3(b) (20, 40, 60, 80, 100 and 120 keV, respectively). The x-ray tube voltage was set at 120 kVp 49 [ Fig. 3(a)]. The distance between the x-ray source and the isocenter was 400 mm; and the distance between the x-ray source and the detector was 1100 mm. A total of 720 projection views were obtained over the 360-degree scanning range.

2.C.2 | Experimental study
The experimental study utilized the head part of the Rando phantom. However, unlike the simulation study, avoiding metal implants along the beam direction was difficult as gantry tilt is available only in the anterosuperior to posteroinferior direction and in the posterosuperior to anteroinferior direction. Therefore, the phantom was scanned in an oblique direction, being rotated by 15°along the vertical axis as shown in Fig. 5. The x-ray tube voltage was again set at 120 kVp; the distance between the x-ray source and the isocenter was 605 mm; and the distance between the x-ray source and the detector was 1062 mm.  Regions-of-interest (ROIs) that we used for quantitative evaluation are indicated in Fig. 6. The regions of soft tissue were subjected to the assessments, as determined by mean absolute percentage error (  whereas the dual energy-based method usually requires a particular system specification, for example, two x-ray tubes producing different voltages or a single x-ray tube with fast voltage switching.

3.B | Experimental study
One drawback of our proposed method is its sensitivity to noise.
We demonstrated that denoising in the modified SSIM was successful. Although a simple Gaussian smoothing was sufficient in this study, improved techniques, such as adaptive filtering, would be desirable as these methods can better conserve the structure and edge information of the object. Because complementary information depends on the tilt angle and direction, further studies are needed to optimize the tilt angle and direction in a given clinical situation.
Application of the scout image acquired before obtaining the patient's CT image may provide a clue for such optimization.

CONFLI CTS OF INTEREST
No conflicts of interest.

APPENDIX
Here, we demonstrate the problem of using original SSIM for calculating correlations between CT images and artifact superposition