Unintended Consequences: Review of New Artifacts Introduced by Iterative Reconstruction CT Metal Artifact Reduction in Spine Imaging

SUMMARY: Metal hardware serves as a common artifact source in spine CT imaging in the form of beam-hardening, photon starvation, and streaking. Postprocessing metal artifact reduction techniques have been developed to decrease these artifacts, which has been proved to improve visualization of soft-tissue structures and increase diagnostic confidence. However, metal artifact reduction reconstruction introduces its own novel artifacts that can mimic pathology.

M etal artifacts are an obstacle to obtaining high-fidelity CT images in postoperative spine imaging, which is increasingly problematic as the number of instrumented spinal fusions grows. There were approximately 463,200 spinal fusions in 2014 in the United States, 1 a 12% increase compared with 2011 2 and nearly triple the amount from 1998. 3 Unfortunately, up to onethird of patients experience no improvement or worsening of symptoms following surgical intervention 4 and require appropriate spine imaging. [4][5][6][7] The main artifacts introduced by metallic spinal implants include beam-hardening, photon starvation, and streaking, 8 which diminish overall image quality and impair the identification of pathology. 9,10 Metal artifact reduction (MAR; GE Healthcare, Milwaukee, Wisconsin) postprocessing techniques have been developed 11 to recover image quality and detail in affected areas and to diminish the artifacts themselves. 12 Unfortunately, MAR introduces different artifacts that can mimic pathology, which radiologists need to recognize. 12,13 The purpose of this article was to review MARrelated artifacts seen on GE Healthcare scanners. While only 1 manufacturer is included in this article, many of the concepts are applicable to different vendor products that use similar techniques.

MAR Technique and Its Benefits
MAR is an automated postprocessing, projection-based technique developed by GE Healthcare that addresses metal artifacts in 3 stages. First, metal artifacts are identified in the source tomogram projection using a density threshold. Second, the data lost from metal artifacts are reconstructed into an inpainted projection with corrected data, which is generated by estimation from artifact-free areas. Finally, the corrected projection is formed by combining the inpainted projection and the original projection, which reveal details of structures obscured by artifacts. 14, 15 MAR reconstruction has been shown to significantly improve visualization of obscured soft-tissue structures and diagnostic confidence compared with the standard weighted filtered backprojection reconstruction method. 12 Paraspinal soft tissues adjacent to the fusion hardware are obscured on traditional images, while the MAR image allows adequate visualization of the underlying structures (Fig 1). Despite the clear benefits of applying MAR to decrease metal artifacts, MAR introduces its own unique artifacts.

MAR-Related Artifacts and Limitations
MAR-induced artifacts have been described in the hip and elbow, 15 but there are limited descriptions of these artifacts in the spine 12,16 where MAR artifacts frequently mimic pathology. The primary artifacts include the following: perihardware lucency, pedicle screw lucency, factitious subarachnoid material on myelography, and misrepresentation of intraosseous cement.
Perihardware Lucency. Perihardware lucency (Fig 2A, white arrow) is concerning because it usually signifies loosening of the hardware; however, this is not visualized in the non-MAR images ( Fig 2B). More troublesome is that these artifacts persist in multiple planes as shown in the axial plane (Fig 2C, -D).
Pedicle Screw Lucency. Pedicle screw lucency occurs when the implanted hardware fatigues and fractures ( Fig  3A, white arrow), which is not seen on the non-MAR image (Fig 3B). This artifact is not uniform throughout the MAR series, and adjacent images demonstrate that adjacent screws are unaffected.
Subarachnoid Material on Myelography. Factitious subarachnoid material can be seen on myelograms when MAR is applied, particularly in areas of concentrated intrathecal contrast. There is hypodense material in the dorsal spinal canal (Fig 4A, white  arrow), which is not seen on the non-MAR image (Fig 4B). This artifact could be mistaken for arachnoiditis, tumor, or layering debris and may lead to unnecessary further work-up.
Intraosseous Cement Abnormal Distribution. Postvertebroplasty cement configuration can appear irregular and fragmented with indistinct margins (Fig 5A, white arrow) and an internal heterogeneous pattern of vertebral body filling; however, in the non-MAR image (Fig 5B), the bone cement has a normal uniform distribution with clearly defined margins.

CONCLUSIONS
MAR is a useful imaging reconstruction technique that can minimize metal artifacts, thus improving softtissue visualization and diagnostic confidence in the setting of spinal hardware; 12 however, it is critical to understand the generated artifacts to render a correct interpretation. One limitation of this article is that we exclusively used GE Healthcare scanners. While MAR techniques of different vendors may have overlap in method (ie, projection modification is used by GE Healthcare and Philips Healthcare), each vendor ultimately has its own proprietary algorithm. 17 All can potentially  simulate undersampling due to oversmoothing by the MAR algorithm in areas where dense objects interface with bone, as shown in previous phantom modeling demonstrating distortion of hardware. 17 At our institution, we include a set of standard images along with the MAR postprocessed images (from a single CT acquisition) on the PACS to allow confident identification of MAR artifacts and true pathology or postoperative complication. Given the widespread use of MAR techniques and the implication of misinterpreting artifacts, it is crucial that radiologists recognize these artifacts exist.