Evaluation of projection‐ and dual‐energy‐based methods for metal artifact reduction in CT using a phantom study

Abstract Objectives Both projection and dual‐energy (DE)‐based methods have been used for metal artifact reduction (MAR) in CT. The two methods can also be combined. The purpose of this work was to evaluate these three MAR methods using phantom experiments for five types of metal implants. Materials and Methods Five phantoms representing spine, dental, hip, shoulder, and knee were constructed with metal implants. These phantoms were scanned using both single‐energy (SE) and DE protocols with matched radiation output. The SE data were processed using a projection‐based MAR (iMAR, Siemens) algorithm, while the DE data were processed to generate virtual monochromatic images at high keV (Mono+, Siemens). In addition, the DE images after iMAR were used to generate Mono+ images (DE iMAR Mono+). Artifacts were quantitatively evaluated using CT numbers at different regions of interest. Iodine contrast‐to‐noise ratio (CNR) was evaluated in the spine phantom. Three musculoskeletal radiologists and two neuro‐radiologists independently ranked the artifact reduction. Results The DE Mono+ at high keV resulted in reduced artifacts but also lower iodine CNR. The iMAR method alone caused missing tissue artifacts in dental phantom. DE iMAR Mono+ caused wrong CT numbers in close proximity to the metal prostheses in knee and hip phantoms. All musculoskeletal radiologists ranked SE iMAR > DE iMAR Mono+ > DE Mono+ for knee and hip, while DE iMAR Mono+ > SE iMAR > DE Mono+ for shoulder. Both neuro‐radiologists ranked DE iMAR Mono+ > DE Mono+ > SE iMAR for spine and DE Mono+ > DE iMAR Mono+ > SE iMAR for dental. Conclusions The SE iMAR was the best choice for the hip and knee prostheses, while DE Mono+ at high keV was best for dental implants and DE iMAR Mono+ was best for spine and shoulder prostheses. Artifacts were also introduced by MAR algorithms.


| INTRODUCTION
The prevalence of metal arthroplasty and implants has been increasing in the United States. 1-3 More than 7 million Americans were reported to have total knee arthroplasty or total hip arthroplasty in 2014. 1 Total shoulder arthroplasty procedures increased by about 5% annually between 1993 and 2007 and were predicted to further increase. 3 A wide range of spine instrumentation has been used for various clinical indications such as trauma, tumors, and degenerative disk disease. 4 Dental implants have also become ubiquitous nowadays both for health and cosmetic reasons. 5 Medical imaging procedures, including computed tomography (CT), are frequently performed in all of these populations for planning, treatment guidance, or diagnosis. For example, the degree of osseous fusion is often evaluated using CT for imaging bony details in the spine. 4 Unfortunately, metal prostheses and implants are often associated with substantial artifacts in medical imaging. In CT, the metals can cause severe artifacts in a form of streaking or shadowing throughout the images due to a number of issues, including beam hardening, photon starvation, noise, scattering, and nonlinear partial volume effect. The appearance of these artifacts varies significantly, depending on metal composition, size, and orientation, as well as CT acquisition parameters. These artifacts often significantly undermine radiologist diagnostic performance and confidence.
Extensive research efforts have been devoted to metal artifact reduction (MAR) in CT. [6][7][8][9][10][11][12] One major theme is to identify the metal implant-corrupted region in the projection data and replace the affected data using different inpainting/interpolation methods.
Another is to use statistics-based and/or model-based iterative reconstruction by segmenting the metal implant-corrupted region and utilize prior knowledge of the imaging physics, system geometry, and noise properties to improve reconstruction quality. All major manufacturers have developed their own proprietary metal artifact reduction techniques. For example, one of the major CT manufacturers combines normalized metal artifact reduction (NMAR) and frequency-split metal artifact reduction (FSMAR) strategies working in projection and image spaces in an iterative fashion (iMAR, Siemens Healthcare, Germany). 7,13 NMAR involves metal segmentation, computation and forward-projection of artifact-free prior images, normalization of the original sinogram by the prior sinogram, and interpolation. Therefore, it avoids direct interpolation which could generate additional artifacts due to unsmooth transition between original and direct-interpolated projection data. FSMAR makes uses of the high-frequency information of the original images and low-frequency information of NMAR-corrected images. Consequently, a spatially weighted sum is generated in order to maintain edges and fine anatomical structures as well as low level of noise. The scope of this work will focus on the different MAR methods from this specific manufacturer. Some technical details of the MAR methods from other manufacturers can be found in a few previous publications, such as Huang et al. and Andersson et al. 14,15 Clinical performances using these commercially available techniques vary to some extent, especially for different metal implant types. [14][15][16][17][18][19][20] In addition to the abovementioned MAR methods, virtual monochromatic images generated from dual-energy (DE) CT is also being used for metal artifact reduction. [21][22][23][24] Low and high tube voltage scans from DE CT contain different spectral information and allow the synthesis of virtual monochromatic images through basis material decomposition. 25 Monochromatic images at low energy provide better iodine contrast-to-noise ratio, while high-energy images can minimize metal artifacts because the appropriately chosen weighting factor could cancel out some of the artifacts between the two basis material images. With virtual monochromatic images, beam-hardening artifacts can be reduced while other factors such as scatter and photon starvation could still remain to affect the images.
This approach was shown to provide promising metal artifact reduction effects for different metal prostheses or implants. 24,[26][27][28][29] Recently, the combination of projection-based and DE-based methods becomes available. However, the efficacy of the combination method has not been reported in the literature. Therefore, the aim of this study was to systematically evaluate three MAR methods (iMAR, virtual monochromatic imaging, and the combination of the two methods), in comparison to single energy without iMAR and DE linearly mixed images for the task of metal artifact reduction using a phantom study.

2.A | Phantoms and experimental setup
Five phantoms (spine, dental, hip, shoulder, and knee) were constructed to evaluate five popular types of metal arthroplasty and implants:

2.B | CT scans
Phantoms were scanned on a dual-source CT scanner (SOMATOM Definition Flash, Siemens Healthcare, Erlangen, Germany) using both single-energy (SE) and DE protocols with the same volume CT dose index (CTDIvol) for each phantom type. Scan parameters were kept as close as possible to our routine clinical exam protocol with reconstruction kernels matched between SE and DE scans. The detailed scan and reconstruction parameters can be found in Table 1. For CNR comparison in the spine phantom, additional SE images at tube potentials of 80, 100, and 140 kV were also acquired with the same CTDIvol. All reconstructions were performed using an iterative reconstruction (SAFIRE, Sinogram-Affirmed Iterative Reconstruction, Siemens Healthcare, Erlangen, Germany) at a strength level of 3.

2.C | Data processing
All SE data were processed using iMAR with respective implant setting, that is, spine, dental, shoulder, hip, and knee settings.
Monochromatic images at 130 keV were synthesized from DE CT using commercially available software (Mono+, Syngo.Via, Siemens Healthcare, Erlangen, Germany). 30 In addition to iMAR and Mono+, a combined method was also used to reduce metal artifacts. In this combined method, the DE raw data were processed first using iMAR for both the low-and high-kV scans, and then the reconstructed low-and high-kV images were loaded to the Mono+ software to generate the Mono+ images at 130 keV. This combined method is referred to as DE iMAR Mono+. Therefore, each phantom generated a total of five sets of images, including three sets with artifact reduction (SE iMAR, DE Mono+, DE iMAR Mono+), DE images mixed with a ratio of 0.5 between the low-and high-kV images (DE mixed) and SE images (Fig. 1).   either 32 9 0.6 mm or 40 9 0.6 mm, which could lead to slight differences in scattering and noise. Third, although we tried our best to obtain a variety of metal implants, only limited numbers were available to us for our study. In addition, the exact composition of these implants was not available. Literature provides some insights 33-35 ; however, there are still a lot of relevant materials including metal alloys, ceramics, and polymers, with a wide range of densities.

2.D | Data analysis
Finally, because of certain limitations of the phantoms, that is, pedicle screws cannot be removed from the spine phantom and metal fillings cannot be removed from the dental phantom, we could not acquire metal artifact-absent reference images for artifact quantification. Evaluations of patient data with metal implants or prostheses is needed to confirm these phantom study results.