Development of a stereoscopic CT metal artifact management algorithm using gantry angle tilts for head and neck patients

Abstract Dental amalgams are a common source of artifacts in head and neck (HN) images. Commercial artifact reduction techniques have been offered, but are substantially ineffectual at reducing artifacts from dental amalgams, can produce additional artifacts, provide inaccurate HU information, or require extensive computation time, and thus offer limited clinically utility. The goal of this work was to define and validate a novel algorithm and provide a phantom‐based testing as proof of principle. An initial clinical comparison to a vendor's current solution was also performed. The algorithm uses two‐angled CT scans in order to generate a single image set with minimal artifacts posterior to the metal implants. The algorithm was evaluated using a phantom simulating a HN patient with dental fillings. Baseline (no artifacts) geometrical measurements of the phantom were taken in the anterior–posterior, left–right, and superior–inferior directions and compared to the metal‐corrected images using our algorithm to evaluate possible distortion from application of the algorithm. Mean HU numbers were also compared between the baseline scan and corrected image sets. A similar analysis was performed on the vendor's algorithm for comparison. The algorithm developed in this work successfully preserved the image geometry and HU and corrected the CT metal artifacts in the region posterior to the metal. The average total distortion for all gantry angles in the AP, LR, and SI directions was 0.17, 0.12, and 0.14 mm, respectively. The HU measurements showed significant consistency throughout the different reconstructed images when compared to the baseline image sets. The vendor's algorithm also showed no geometrical distortion but performed inferiorly in the HU number analysis compared to our technique. Our novel metal artifact management algorithm, using CT gantry angle tilts, provides a promising technique for clinical management of metal artifacts from dental amalgam.


| INTRODUCTION
Computed tomography (CT) imaging artifacts are discrepancies between Hounsfield unit (HU) values and actual linear attenuation coefficients of the object. These can pose a problem for physicians who are completing a diagnosis or attempting to identify or delineate the extent of disease. In the reconstruction of the CT images, dense objects such as bone and metal are a common source of artifacts through beam hardening and photon starvation. Beam hardening is caused by the presence of dense (and high atomic number) structures in the beam path. The x-ray spectrum undergoes an upward shift in average energy due to the preferential attenuation of lower-energy photons. CT reconstruction algorithms attempt to correct for beam hardening but are optimized for human tissues and cannot fully address highly attenuating materials such as metals. 1 Photon starvation occurs when these highly attenuating materials cause the exiting x-rays to have a low photon flux on the detectors.
Consequently, the combination of beam hardening and photon starvation produces streaking artifacts in the reconstruction and can affect the image severely.
The most common artifacts present in head and neck (HN) cancer patients are the ones caused by the presence of high atomic number materials in the image, such as the ones originated by dental fillings.
Dental filling metal amalgam artifacts can obscure the visualization of tumors in the oral cavity and oropharynx. This obscuring of the anatomy can lead to poor visualization of tissues and therefore improper definition of the target, potentially providing suboptimal management of the disease, particularly including radiotherapy quality. Studies have shown that the presence of dental artifacts can in fact increase the inter-observer contouring variability of HN tumors. 2 Aside from the difficulty in visualization of tumors and in definition of planning target volumes and organs at risk (OARs), metal artifacts will alter the true HU values in the affected voxels which negatively impacts the quality of radiotherapy in such areas. Kim et al. and Mail et al. have demonstrated that such artifacts result in increased dose heterogeneity and reduced target coverage. 3,4 In photon therapy, calculation errors were found to exceed 10% in an oral cavity clinical target volume when fillings are present, compared with 3% when no metal was present, emphasizing the potential severity of metal artifacts on dose calculations. 5 The consequences for dose calculation accuracy are also particularly relevant in proton therapy because of the strong dependency between a correct relative linear stopping power prediction and accurate representation of HU values. [6][7][8][9] Proton treatment plans could display erroneous beam ranges and dose distributions when artifacts are present.
Several solutions for metal artifact reduction have been proposed, but many are impractical or not clinically feasible and therefore are not extensively adopted. Newhauser et al. accomplished a significant range uncertainty improvement in proton treatment planning with the use of megavoltage (MVCT) CT images. 10 [13][14][15][16] and can generally be divided into two groups: iterative reconstruction algorithms and projection completions methods. The former approach starts from an initial guess image, re-projects the image to the sinogram space, compares it to the original projections to generate a correction, applies that correction, and repeats that process until the difference between the images is minimized. This approach is superior at handling metal artifacts but it requires extensive computation time, making the technique clinically unfeasible. The latter approach works by replacing the corrupted projections in the sinogram space with interpolated data from regions of the sinogram unaffected by the metal. The estimation of the missing raw data values will determine how successful the algorithm is. Sharp transitions between the original projections and the interpolated ones cause additional artifacts. Moreover, the estimation of raw data values creates blurring in the final image due to data loss near the metal edges, which is not recoverable through the estimation of values. Despite the creation of additional artifacts and direct interpolation of HU information, projection completion methods gained more popularity. However, a recent study of three current commercially available artifact reduction methods concluded that they were generally not successful at reducing artifacts specifically caused by dental fillings. 17 Indeed, particularly for dental artifacts, that study found that the commercial solutions had either a minimal effect or actually made the artifacts worse. Other post-processing metal artifact reduction algorithms have been published but have not found clinical acceptance. [18][19][20] Despite several publications of metal artifact reduction algorithms over the past two decades, there remains an evident need for better metal artifact management in highly heterogeneous sites, such as the HN. 5,17,21 To address the need for better metal artifact management, we developed an algorithm that is not based on direct interpolation methods and therefore will not require the removal, replacement, and consequential loss of data points. In addition, the algorithm will not be system specific and thus could be used with BRANCO ET AL.
| 121 any CT scanner that allows for gantry tilts, which is a feature offered on scanners from all major CT manufacturers. In this work, we will introduce the artifact management algorithm and provide testing on a geometrical phantom as a proof of principle. An initial clinical comparison to a vendor's current algorithm solution will also be presented.

2.A | Algorithm
Similar to the concept of stereoscopic imaging, the algorithm developed in this work makes use of two angled CT scans to generate one artifact-reduced image set. The issue with traditional 0°scans on patients with HN disease who have dental fillings or implants is that the artifact-compromised slices are located where typical HN disease is located (Fig. 1), posterior to the oral cavity.
The goal of this algorithm was to use two-angled CT scans to reconstruct an image where the posterior region can display the accurate HU information without the need for the widely used The framework for our new algorithm is divided into two main steps.
Step 1 is responsible for untilting the images that were acquired at an angle, so that they appear as regular axial slices and can then be used in step 2.
Step 2 uses those untilted images to form an image with the metal artifacts greatly reduced. The complete artifact reduction routine depends on the number of images per scan but on average takes less than 1 min to complete. F I G . 1. Diagram representing a patient with head and neck disease with the range of CT slices affected by the dental work. The red circle shows the region of typical head and neck disease that gets affected by artifacts resulting from metal in the mouth.

2.B | Evaluation
The routine described in the previous section was tested on a geometrical phantom simulating a HN cancer patient with dental fillings. The      The HU measurements also showed no correlation with varying gantry angles and are shown in Fig. 7. Linear regression lines were also fitted for each material, and the same regression analysis was performed as described above. Similar to the distortion measurements, all of the material slopes' P-values were above significance level, indicating no pattern correlating HU with gantry angle tilt.
Additionally, for each material, the HU values measured were statistically consistent (within the 95% CI) with the true (untilted) HU value. In the quantitative comparison between both techniques, it was discovered that SmartMAR corrected images showed no relevant geometrical distortion. The average total distortions for all plugs and phantom in the AP, LR, and SI directions were very small, 0.03, 0.05, and −0.3 mm, respectively. Table 3 shows the HU numbers collected and their standard deviations (SD) inside the fixed ROIs for all scans. Table 4 shows the HU number differences between the relevant scans; metal uncorrected and respective baseline, and metal corrected (SmartMAR and our Technique) and respective baselines. Ideally the HU number difference should be close to 0, indicating no difference between the baseline scan and the corrected image set in question. However, the metal uncorrected scans are filled with dark and light streaks hence showing large HU differences. The streaking is also indirectly represented by the SD inside the ROIs (Table 3); the uncorrected image sets showed large variations in HU numbers, hence displaying large SD values. It is possible to notice that, for most plugs (blue water, Techtron HPV, solid water, and phantom), SmartMAR provided some improvement in the HU accuracy with HU differences closer to 0 then the metal uncorrected images. However, the HU number values were still notably different from what they were expected to be (baseline). It is important to notice that the PBT and cork plugs' HU values were worsened by SmartMAR's algorithm. In contrast, our technique showed very small differences F I G . 6. Results for distortion measurements showing each difference obtained between the measurements of the different plugs and phantom done with the gantry at 0°with no metal teeth present (baseline) and with metal teeth present for the six different CT gantry angles in the AP (a), LR (b), and SI (c) directions.  (Table 3).

| DISCUSSION
The algorithm generated in this work was successful at eliminating metal artifacts created by dental amalgam in the posterior region of the image. It is possible to see that as the CT gantry angle increases, the posterior region becomes clearer of metal-affected pixels, leading to better visualization of the structures in the phantom. As a consequence of the combination of two angled scans, the artifacts extend to regions that were previously unaffected, such as the nose and chin. However, as previously mentioned, those areas do not normally contain disease or OARs.
Several metal artifact reduction methods have been proposed and are currently available to the community. These methods include the use of MVCT, dual-energy CT, magnetic resonance imaging, additional CT scans, and the complete removal of the dental work.
However, each has important limitations and hence lacks wide clinical acceptance. A common technique in radiation oncology is to manually override HU values, but this has major drawbacks in that anatomy is still obscured and is now assumed to be homogeneous. The presented methodology in this manuscript was the introduction of a stereoscopic solution to reduce metal artifacts present in patients' HN CT images through the use of CT gantry angles. Future work will expand the image quality and robustness study (effects of head tilt, slice thickness, etc) comparing this technique to all major vendors' current metal artifact reduction algorithms with the use of a HN anthropomorphic phantom. We will also further investigate the performance of the technique in the context of radiation therapy and treatment planning system dose calculations. A dosimetric analysis will be performed on the anthropomorphic phantom to show the advantages of the algorithm developed here over the other approaches currently in use. Proton treatment planning dose calculations and proton beam differences will be of particular interest because of their large dependency on HU accuracy and robustness.

| CONCLUSION
A stereoscopic metal artifact management algorithm was developed using CT gantry angle tilts and evaluated in a geometrical HN phantom. The algorithm developed here offered the improvement of not requiring the replacement of deleted metal thresholded data with artificially interpolated data. In addition, it used accurate HU data obtained from two different scans and was divided into two parts: the first included the untilting and correction of the angled image set, and the second involved the removal of the metal artifacts in the posterior region. Unlike other existing algorithms, this algorithm is independent of the CT scanner provider and therefore can be used in any scanner that allows for gantry tilts. Also, our technique is applied in the image space, and therefore does not require the need to acquire and manipulate the proprietary raw data from vendors. The images showed the successful removal of the artifacts present in the posterior region on the phantom, allowing for much better visualization of the structures. The quantitative analysis of the algorithm performance showed that it presents artifact corrected images with no geometrical distortion and with HU number accuracy when compared to the baseline. In addition, our technique outperformed a commercially used algorithm, SmartMAR, in providing artifact-free images with better HU agreement with the metal-free baseline. Future work will be done to further expand the image quality analysis and robustness among all major vendors' solutions, and to evaluate treatment planning dosimetry, specifically applied to proton therapy since proton treatment quality and robustness are highly dependent on HU accuracy.

ACKNOWLEDGMENTS
The authors thank the MD Anderson's Department of Scientific Publications for the careful reading of the manuscript.

CONF LICT OF I NTERESTS
The authors have no relevant conflicts of interest to disclose.