Advances in Bone Joint Imaging-Metal Artifact Reduction

Numerous types of metal implants have been introduced in orthopedic surgery and are used in everyday practice. To precisely evaluate the postoperative condition of arthroplasty or trauma surgery, periprosthetic infection, and the loosening of implants, it is important to reduce artifacts induced by metal implants. In this review, we focused on technical advances in metal artifact reduction using digital tomosynthesis, computed tomography, and magnetic resonance imaging. We discussed new developments in diagnostic imaging methods and the continuous introduction of novel technologies to reduce metal artifacts; however, these innovations have not yet completely removed metal artifacts. Different algorithms need to be selected depending on the size, shape, material and implanted body parts of an implant. Future advances in metal artifact reduction algorithms and techniques and the development of new sequences may enable further reductions in metal artifacts even on original images taken previously. Moreover, the combination of different imaging modalities may contribute to further reductions in metal artifacts. Clinicians must constantly update their knowledge and work closely with radiologists to select the best diagnostic imaging method for each metal implant.


Introduction
Advances in bone joint imaging in the past few decades have contributed to accurate diagnoses and the selection of appropriate treatments, particularly in the field of orthopedic surgery. Since many types of metal implants have been used in orthopedic surgery, reductions in metal artifacts are highly favorable for assessing the postoperative condition of arthroplasty or trauma surgery, periprosthetic infection, and the loosening of implants. In this review, we focus on technical advances in metal artifact reduction, which is essential in the orthopedic field to allow for precise evaluations of the conditions around metal implants. The purpose of this review was to discuss recent advances in metal artifact reduction in diagnostic imaging for orthopedic surgery as well as future perspectives.

Materials and Methods
The modalities we discuss are digital tomosynthesis (DT), computed tomography (CT), and magnetic resonance imaging (MRI). We searched for articles using Google scholar in the period of 2016-2022. Articles on DT were searched for using the keywords "digital tomosynthesis; metal artifact reduction; musculoskeletal", those on CT using "CT; dualenergy CT; metal artifact reduction; musculoskeletal", and those on MRI using "MRI; metal artifact reduction; musculoskeletal; sensing; sequences". There were 269, 357, and 2250 articles, respectively, for each modality. We excluded case reports and articles that did not describe the method for metal artifact reduction and reviewed those with specific . DT acquires image data by a single sweep of an X-ray tube, with a range of 8 to 40°, whereas that of cone beam computed tomography is 360°. The X-ray is detected by digital flat panel detectors (FPD), which are either in the same position or may move in the opposite direction to the X-ray tube depending on the tables. The important reconstruction algorithm for DT is an iterative reconstruction (IR) [7] because it is often used with a process called tomosynthesis with metal artifact reduc-tion (TMAR), which is introduced using metal extraction and ordered subset-expectation maximization reconstruction [12]. This algorithm improves image quality and in-depth spatial resolution. Projection images were initially separated into metal and metal-free images, and iteratively reconstructed to reduce metal artifacts; these two images were then fused [5] (Figure 2). TMAR processing is better than radiography at reducing overlapping structures, with markedly lower radiation exposure doses and fewer artifacts than those with CT [13,14]. The important reconstruction algorithm for DT is an iterative reconstruction (IR) [7] because it is often used with a process called tomosynthesis with metal artifact reduction (TMAR), which is introduced using metal extraction and ordered subset-expectation maximization reconstruction [12]. This algorithm improves image quality and in-depth spatial resolution. Projection images were initially separated into metal and metal-free images, and iteratively reconstructed to reduce metal artifacts; these two images were then fused [5] (Figure 2). TMAR processing is better than radiography at reducing overlapping structures, with markedly lower radiation exposure doses and fewer artifacts than those with CT [13,14]. Brief image of tomosynthesis with metal artifact reduction (TMAR). Projection images were initially separated into metal and metal-free images. These images were iteratively reconstructed to reduce metal artifacts and were then fused to create TMAR images.
One of the major advantages of DT is its high in-section spatial resolution. The TMAR algorithm also improves depth resolution when metal objects are present. However, DT does not allow the production of multiplanar reconstructions, as in CT, due to the limited number of projections and the acquisition angle [7].
Even though DT reduces metal artifacts, some new artifacts are generated, such as undershooting, blurring-ripple, and ghost artifacts [7]. TMAR reduces these artifacts and improves the quality of the obtained image and in-depth spatial resolution [13,15], while simultaneously increasing the signal of metal objects in its peripheral part [7]. Inversion of the gray scale may improve the reading comfort of the tomosynthesis images; however, its diagnostic value remains unchanged [16].
In the diagnosis of fractures, DT is considered to be superior to radiography, but inferior to CT [7,[17][18][19]. Its performance is optimal when the imaging plane is best suited to the analysis of target structures, such as the femoral neck and tibial plateau [7]. Tang et al. compared the sensitivity and specificity of DT with TMAR against radiography and conventional CT for detecting implant loosening in total hip arthroplasty (THA), namely, radiolucent lines narrower than 2 mm surrounding cementless femoral stems, in four cadavers [20]. The overall sensitivities of DT, radiography, and CT were 63.3, 20.5, and 50.2%, respectively. The sensitivity of DT was significantly higher than those of the other two modalities. They concluded that DT has the potential to increase the diagnostic accuracy of early prosthetic loosening after cementless THA in clinical practice.
In a cohort study of 48 patients, Tang et al. reported that the diagnostic accuracy of DT with TMAR to assess the fixation stability of cementless THA was significantly higher than those of plain radiography and CT for both the femoral and acetabular sides [1]. They assessed obtained images with following diagnostic criteria; loosening, not loosening (definite osteointegration fixation), and possible loosening (not sure). They confirmed fixation stability with two major clinical criteria as a reference standard, intraoperative Brief image of tomosynthesis with metal artifact reduction (TMAR). Projection images were initially separated into metal and metal-free images. These images were iteratively reconstructed to reduce metal artifacts and were then fused to create TMAR images.
One of the major advantages of DT is its high in-section spatial resolution. The TMAR algorithm also improves depth resolution when metal objects are present. However, DT does not allow the production of multiplanar reconstructions, as in CT, due to the limited number of projections and the acquisition angle [7].
Even though DT reduces metal artifacts, some new artifacts are generated, such as undershooting, blurring-ripple, and ghost artifacts [7]. TMAR reduces these artifacts and improves the quality of the obtained image and in-depth spatial resolution [13,15], while simultaneously increasing the signal of metal objects in its peripheral part [7]. Inversion of the gray scale may improve the reading comfort of the tomosynthesis images; however, its diagnostic value remains unchanged [16].
In the diagnosis of fractures, DT is considered to be superior to radiography, but inferior to CT [7,[17][18][19]. Its performance is optimal when the imaging plane is best suited to the analysis of target structures, such as the femoral neck and tibial plateau [7]. Tang et al. compared the sensitivity and specificity of DT with TMAR against radiography and conventional CT for detecting implant loosening in total hip arthroplasty (THA), namely, radiolucent lines narrower than 2 mm surrounding cementless femoral stems, in four cadavers [20]. The overall sensitivities of DT, radiography, and CT were 63.3, 20.5, and 50.2%, respectively. The sensitivity of DT was significantly higher than those of the other two modalities. They concluded that DT has the potential to increase the diagnostic accuracy of early prosthetic loosening after cementless THA in clinical practice.
In a cohort study of 48 patients, Tang et al. reported that the diagnostic accuracy of DT with TMAR to assess the fixation stability of cementless THA was significantly higher than those of plain radiography and CT for both the femoral and acetabular sides [1]. They assessed obtained images with following diagnostic criteria; loosening, not loosening (definite osteointegration fixation), and possible loosening (not sure). They confirmed fixation stability with two major clinical criteria as a reference standard, intraoperative mechanical tests and postoperative retrieval findings. They compared image findings and clinical standards, and categorized them into 'accurate', 'wrong' and 'not sure' groups. They defined diagnostic accuracy as the rate of the 'accurate' group for each imaging examination. Diagnostic accuracies for the femoral stem and acetabular cup were 82.6 and 84.5%, 44.6 and 67.3%, and 39.6 and 74.6% with DT, radiography, and CT, respectively. They concluded that DT with TMAR increased the diagnostic accuracy of assessments of the fixation stability of cementless THA by minimizing metal artifacts in the border between the implant and bone and clearly depicting peri-implant trabecular structures. Guo et al. reported similar findings for the usefulness of DT with TMAR in evaluations of periprosthetic conditions [21]. They investigated the osteointegration of 24 patients who underwent revision cementless THA. All the patients underwent radiography, DT with TMAR, and CT prior to the surgery. Evidence of osteointegration in retrieved prostheses was used as the reference standard. They evaluated 13 femoral stems and 14 acetabular components. Sensitivities on the femoral side for radiography, DT with TMAR and CT were 50.4%, 73.8% and 36.4%, respectively. Sensitivities on the acetabular side for radiography, DT with TMAR and CT were 45.9%, 60.2% and 45.1%, respectively. Accordingly, specificities on the femoral sides were 87.8%, 94.3%, and 90.9%, and on the acetabular sides were 66.4%, 86.4% and 73.5%, respectively.
On the other hand, Gillet et al. demonstrated that the diagnostic performance of DT in hip prosthetic loosening was similar to that of radiographs, and its sensitivity was lower than that of CT with the metal artifact reduction algorithm (MAR) [22]. They investigated 49 patients with painful hip prostheses. Among them, 21 cases were confirmed as prosthetic loosening by surgery. Sensitivities for radiography, DT and CT with MAR were 33.3-51.5%, 39.9-45.4%, and 84.5%, respectively. Specificities were 96.9-100%, 98.5-100%, and 95.4-96.9%, respectively. Since the interobserver agreement of DT was higher than that of radiography (0.53 vs. 0.39), particularly for the acetabular component, they recommended the use of DT in combination with conventional radiography. In cases with an inconclusive initial evaluation with radiography and digital tomosynthesis, additional evaluations by CT with MAR were recommended [22].
Metal artifact reduction by DT is advantageous for assessing bone formation after various surgical procedures other than arthroplasties, particularly in the presence of metal implants. Toyooka et al. investigated 27 patients and reported that the evaluation of bone integration after anterior cruciate ligament reconstruction of the knee with DT was equivalent to that with CT within 15% of diagnostic error. DT showed a sensitivity of 79-96%, specificity of 64-100% and diagnostic accuracy of 81-96% [23]. Ishibashi et al. demonstrated the applicability of DT to assessments of postoperative bone formation in the opening gap after open wedge high tibial osteotomy (OWHTO). They investigated the correlation between gap filling value (GFV) and modified van Hemert's score (MVHS)for assessment of bone formation after OWHTO. GFV and MVHS showed a strong correlation (r = 0.630, p < 0.001). The interclass correlation coefficient (ICC) for intraobserver reliability was 0.958 for GFV and 0.978 for MVHS. ICC for interobserver reliability was 0.975 for GFV and 0.950 for MVHS. They concluded that the evaluation of bone formation after OWHTO using DT has high accuracy and reproducibility [24].
DT is also useful in the field of spine surgery. Okano et al. developed a novel pixel selection method for pedicle screws (PS) by DT [25]. They enlarged the obtained images up to 3200 times and observed the luminance of each pixel to identify the peripheral line of the PS. Using the identified peripheral pixels, they identified rod curve lines. From the (x, y) coordinates of ten pixels on the rod curve line, they calculated quadratic regression curves. This method is useful when identifying the same cross-sectional slice from two different datasets obtained separately. The two slices are on the same cross-section if the shapes of the regression curves are the same. Using this method, Mataki et al. evaluate PS loosening in 41 patients with 72 PS by DT. They concluded that DT has the potential to diagnose PS loosening more accurately and quantitatively than conventional modalities [26] (Figure 3).  The most recent advance in DT is the application of a deep learning program to achieve more advanced metal artifact reduction. Gomi et al. reported a novel projectionbased cross-domain learning framework for MAR [27]. They used the novel algorithm and successfully reduced metal artifacts more than with conventional TMAR algorithms. They also succeeded in reducing the radiation dose by 55% [27].
In summary, DT provides higher in-section spatial resolution than plain radiography, with markedly fewer metal artifacts and lower radiation doses than with conventional CT. Even though multiplanar reconstruction is impossible, DT is a useful tool for evaluating periprosthetic conditions after arthroplasty and bone formation after fractures or osteotomies. In cases in which the assessment of soft tissues is not needed and the lesion of interest is limited, DT may outperform CT or MRI after arthrography [15,28] (Table 1).  Figure 3. Comparison among plain radiography, computed tomography (CT), and digital tomosynthesis (DT) in an evaluation of pedicle screw loosening after spinal instrumentation. In DT, radiolucent areas around PS were more evident than in images obtained by the two other modalities (arrows).
The most recent advance in DT is the application of a deep learning program to achieve more advanced metal artifact reduction. Gomi et al. reported a novel projection-based cross-domain learning framework for MAR [27]. They used the novel algorithm and successfully reduced metal artifacts more than with conventional TMAR algorithms. They also succeeded in reducing the radiation dose by 55% [27].
In summary, DT provides higher in-section spatial resolution than plain radiography, with markedly fewer metal artifacts and lower radiation doses than with conventional CT. Even though multiplanar reconstruction is impossible, DT is a useful tool for evaluating periprosthetic conditions after arthroplasty and bone formation after fractures or osteotomies. In cases in which the assessment of soft tissues is not needed and the lesion of interest is limited, DT may outperform CT or MRI after arthrography [15,28] (Table 1).

Computed Tomography
Computed tomography (CT) is commonly used to evaluate complications related to metal implants, such as infection, loosening, fracture, implant failure, particle disease, tumors, and pseudotumors [29]. However, a combination of metal artifacts is a major limitation in evaluations by conventional CT. Examples of artifacts are beam hardening, scattering, photon starvation, and edge effects [5,6,30,31]. These metal artifacts result in the reconstruction of near-metal tissue by corrupted data [31]. The severity of metal artifacts on CT is dependent on the atomic number as well as the shape and size of metal implants. Larger implants and metal with higher atomic numbers result in greater metal artifacts [32].
There are three main strategies to achieve metal artifact reduction: modifying standard acquisition and reconstruction, modifying projection data and/or image data, and applying dual-energy CT (DECT) [31]. Increases in the tube voltage (kV) and tube current (mAs) are traditional techniques to reduce beam hardening and photon starvation in conventional CT [31,33]. A reduction in total collimation (detector width) decreases scatter effects [31]. Instead of using standard filtered back-projection (FBP), changing the reconstruction algorithm to projection interpolation techniques or more advanced IR and model-based iterative reconstruction (MBIR) algorithms reduces the scatter and edge effects [29][30][31]34]. However, these techniques require a higher radiation dose and may result in lower spatial and contrast resolution as well as reconstruction errors [35].
The modification of projection and/or image data includes the use of the metal artifact reduction algorithm (MAR), which reconstructs more accurate images by segmenting and extracting metal artifacts iteratively [31]. MAR significantly reduces metal artifacts, particularly the effects of beam hardening and photon starvation [31,32] (Figure 4). The use of MAR in addition to MBIR was previously reported to reduce the radiation dose by up to 80% from that with conventional FBP [31]. Previous studies demonstrated the effectiveness of MAR at reducing metal artifacts following total hip arthroplasty (THA) [36][37][38][39], total knee arthroplasty (TKA) [30], and different metal fixation implants [40][41][42]. Bolstad et al. concluded that MAR reduced metal artifacts, particularly on implants made of steel and cobalt-chrome [42]. However, it is important to note that the use of MAR may induce new minor artifacts in the tissues inside and proximal to metal objects that are not visible on conventional CT, such as streaks or the disappearance of metal implants. The inaccurate segmentation of data or errors in the estimation of corrupted data has been shown to cause these artifacts [31,42,43].
Tissues and materials have different attenuation properties at different energy levels, and the difference in density between a metal implant and musculoskeletal tissue is the main cause of a metal artifact [30]. Soft tissue, water, and adipose tissue show consistent attenuation levels throughout any X-ray beam energy. However, the attenuation levels for calcium and iodine are higher for a lower beam energy level due to the photoelectric effect [44]. DECT takes advantage of this by acquiring CT attenuation data at two different energy levels, generally 70-80 and 140-150 kV [45,46]. After data acquisition, virtual monoenergetic images (VMI) are reconstructed, which is the main characteristic and advantage of DECT. VMI are reconstructed gray-scale images that virtually simulate appearances, which may be achieved using a true monochromatic X-ray beam [33].  Tissues and materials have different attenuation properties at different energy levels, and the difference in density between a metal implant and musculoskeletal tissue is the main cause of a metal artifact [30]. Soft tissue, water, and adipose tissue show consistent attenuation levels throughout any X-ray beam energy. However, the attenuation levels for calcium and iodine are higher for a lower beam energy level due to the photoelectric effect [44]. DECT takes advantage of this by acquiring CT attenuation data at two different energy levels, generally 70-80 and 140-150 kV [45,46]. After data acquisition, virtual monoenergetic images (VMI) are reconstructed, which is the main characteristic and advantage of DECT. VMI are reconstructed gray-scale images that virtually simulate appearances, which may be achieved using a true monochromatic X-ray beam [33].
VMI are useful because they are reconstructed at arbitrary average energy levels and, thus, contrast optimization and artifact reduction are possible. By making VMI with higher virtual monochromatic energies, the influence of low-energy photons may be reduced, leading to fewer beam-hardening artifacts without increasing the radiation dose [31,[47][48][49]. However, DECT cannot reduce photon starvation, scatter, or edge effects.
There are three main types of postprocessing algorithms for DECT datasets: image optimization algorithms (IOA), differentiation algorithms (DA), and quantification algorithms (QA) [45,46]. IOA typically provides two sets of monochromatic images and a nonlinear blended image of high-and low-energy images. Monochromatic images are obtained at 80 or 100 kV and 140 kV. Low-energy images generally provide high contrast, while high-energy images provide low noise. In DA, specific materials may be subtracted from the data, or two materials may be differentiated by color coding. MAR is considered to be one of the differentiation algorithms. QA is based on the decomposition of three materials and provides color-coded images of certain material in postcontrast examinations [44]. VMI are useful because they are reconstructed at arbitrary average energy levels and, thus, contrast optimization and artifact reduction are possible. By making VMI with higher virtual monochromatic energies, the influence of low-energy photons may be reduced, leading to fewer beam-hardening artifacts without increasing the radiation dose [31,[47][48][49]. However, DECT cannot reduce photon starvation, scatter, or edge effects.
There are three main types of postprocessing algorithms for DECT datasets: image optimization algorithms (IOA), differentiation algorithms (DA), and quantification algorithms (QA) [45,46]. IOA typically provides two sets of monochromatic images and a nonlinear blended image of high-and low-energy images. Monochromatic images are obtained at 80 or 100 kV and 140 kV. Low-energy images generally provide high contrast, while high-energy images provide low noise. In DA, specific materials may be subtracted from the data, or two materials may be differentiated by color coding. MAR is considered to be one of the differentiation algorithms. QA is based on the decomposition of three materials and provides color-coded images of certain material in postcontrast examinations [44].
Previous studies reported that DECT reduced metal artifacts on spine implants, implants for fractures, and hip and knee prostheses [40][41][42][50][51][52]. However, there is no generalized optimal kV for metal implants, which may be attributed to differences in metal alloys, the size, shape, and geometry of the implant, the body region, and the acquisition parameters used [31]. However, most studies reported a range of 110-150 kV, with 130 kV being sufficient for most small implants composed of lightweight alloys [31,50,52,53]. Lee  Donders et al. investigated 41 patients with suspected non-union after fracture surgery using intramedullary nails and plates [55]. They performed DECT at high (130-150) kV and low (70) kV and used Likert scores to evaluate the usefulness. It was concluded that the image quality (1.83 vs. 0.88) and diagnostic confidence were higher (2.37 vs. 1.43) and the false negative rates of non-union were 5% lower with high kV DECT than with low kV DECT. Non-union was confirmed during revision surgery [55]. These findings support the effectiveness of metal artifact reduction by DECT in actual practice.
Other than metal artifact reduction, DECT has been used to detect urate crystals in gout or calcium pyrophosphate dehydrate crystal arthropathy as well as bone marrow edema in trauma or inflammation and also to characterize tendons, ligaments, and intervertebral discs [46]. However, VMI may decrease tissue contrast because low-energy photons increase contrast [32]. In addition, when implants are large and have sharp edges or high-molecularweight metals, DECT artifact reduction is less effective [45].
Barreto et al. compared the effectiveness of MAR, conventional CT images, and images obtained from six cadavers containing metal implants in the head, neck, abdomen, pelvis, and extremities using DECT [43]. Left hip bipolar hemiarthroplasty, TKA, and an implant for anterior cervical disc fusion were included as musculoskeletal implants. The severity of metal artifacts, the visualization of anatomical structures, and the assessors' confidence in diagnostic interpretation were assessed using the original 5 points scale. In all six cadavers, MAR was preferred over conventional CT images, while conventional CT images were preferred over DECT images. Since DECT reduced soft tissue contrast and streak artifacts remained, it was difficult to evaluate the surrounding structures. The cervical spine was the only case in which DECT reduced the severity of metal artifacts and increased the visibility of all structures. They concluded that MAR was more effective at reducing metal artifacts than DECT; however, new minor streak artifacts were observed in MAR images. When the anatomy or implant is relatively small, DECT may be superior to MAR without additional artifacts [43].
Previous studies suggested that VMI by DECT in combination with advanced reconstruction algorithms, such as projection interpolation techniques and MAR, were more likely than VMI alone to reduce beam hardening artifacts [45]. Neuhaus et al. reported that the application of VMI or MAR was not sufficient to reduce severe metal artifacts induced by monolateral or bilateral hip prostheses. They concluded that the combination of VMI and MAR achieved greater reductions in metal artifacts. Additionally, the assessment of adjacent structures such as pelvic organs and bone improved the most with the combination of VMI and MAR [41]. Andersson et al. investigated several MAR and VMI for the metal artifact reduction of bilateral hip prosthesis phantom and showed that the use of VMI alone did not decrease the artifact to the same extent as the combination of VMI and MAR [56]. Furthermore, in the region of interest between bilateral hip prostheses, the artifact even increased with the use of VMI. Bongers et al. compared the metal artifact reduction performance of blended polychromatic images (equivalent to 120 kV) and monochromatic images at 130 kV with and without MAR on 20 patients with hip prostheses and 30 patients with dental implants [57]. They evaluated the artifact reduction rate qualitatively and quantitatively. Compared to conventional CT, DECT reduced by 33% and 8%, MAR reduced by 56% and 71%, and a combination of DECT and MAR reduced by 76% and 76% for hip prosthesis and dental implants, respectively. Long et al. examined twenty patients with instrumented spines using VMI, MAR, and their combination [58]. They used original artifact scores (1 to 5 points) and image quality scores (1 to 4 points), and VMI in combination with MAR showed the best for both scores. duced by 76% and 76% for hip prosthesis and dental implants, respectively. Long et al. examined twenty patients with instrumented spines using VMI, MAR, and their combination [58]. They used original artifact scores (1 to 5 points) and image quality scores (1 to 4 points), and VMI in combination with MAR showed the best for both scores. ICC was 0.779 for the bony structures and 0.892 for the soft tissues. Bongers et al. and Long et al. both concluded that the combination of VMI and MAR achieved the greatest reductions in metal artifacts ( Figure 5).

Figure 5.
Comparison of conventional computed tomography (CT) and images reconstructed with virtual monoenergetic images (VMI), the metal artifact reduction algorithm (MAR), and combination of VMI and MAR. Coronal images of a patient after intramedullary nail fixation of the femur. Image (a) was taken by conventional CT at 120 kV, and shows strong beam hardening, scattering, photon starvation, and edge effects at the junction of the nail and lag screw (arrowheads), and at the tip of distal locking screws (arrows). Image (b) was taken by VMI at 135 kV, and the artifacts are slightly reduced, especially at the tip of distal locking screws. Image (c) was taken with MAR, and the artifact reduction is stronger than VMI, only very small dark streaks both at the junction of the nail and lag screw and at the tip of distal locking screws. Image (d) was taken in combination with VMI at 135 kV and MAR, and there are almost no artifacts present, leading to a better visualization of the outlines of the prosthesis.
Yue et al. evaluated 35 patients with unilateral hip arthroplasty using high (120-140) and low (80-100) kV VMI with and without MAR. Their findings showed that metal artifacts were significantly lower in high kV VMI with MAR, and the image quality of the peri-prosthesis region was better [59]. Chae et al. assessed 57 knees from 36 patients after TKA. They compared images obtained by conventional CT, MAR, VMI, and the Figure 5. Comparison of conventional computed tomography (CT) and images reconstructed with virtual monoenergetic images (VMI), the metal artifact reduction algorithm (MAR), and combination of VMI and MAR. Coronal images of a patient after intramedullary nail fixation of the femur. Image (a) was taken by conventional CT at 120 kV, and shows strong beam hardening, scattering, photon starvation, and edge effects at the junction of the nail and lag screw (arrowheads), and at the tip of distal locking screws (arrows). Image (b) was taken by VMI at 135 kV, and the artifacts are slightly reduced, especially at the tip of distal locking screws. Image (c) was taken with MAR, and the artifact reduction is stronger than VMI, only very small dark streaks both at the junction of the nail and lag screw and at the tip of distal locking screws. Image (d) was taken in combination with VMI at 135 kV and MAR, and there are almost no artifacts present, leading to a better visualization of the outlines of the prosthesis.
Yue et al. evaluated 35 patients with unilateral hip arthroplasty using high (120-140) and low (80-100) kV VMI with and without MAR. Their findings showed that metal artifacts were significantly lower in high kV VMI with MAR, and the image quality of the peri-prosthesis region was better [59]. Chae et al. assessed 57 knees from 36 patients after TKA. They compared images obtained by conventional CT, MAR, VMI, and the combination of MAR and VMI, and found that the combination of MAR and VMI achieved the greatest reductions in metal artifacts and more accurately depicted soft tissues. On the other hand, bony structures were more accurately depicted by MAR alone [60]. These findings suggest the usefulness of the combination of VMI and MAR in evaluations of peri-prosthesis conditions at both the hip and knee.
Similar to DT, deep learning is recently being used in MAR in CT. Park et al. employed U-Net to reduce artifacts in polychromatic CT against beam hardening [61,62]. Zhang et al. concluded that a convolutional neural network was capable of creating priority images with fewer artifacts to correct the regions of a sinogram corrupted by metal artifacts [63]. In combination with MAR, these methods achieved effective metal artifact reduction. However, the processing of new artifacts in reconstructed CT images is still challenging [27]. Recent studies suggested that the use of a generative adversarial network increased the recognition and tissue segmentation accuracy of tissues on CT images [27,64]. In combination with MAR, it may effectively reduce metal artifacts. Future advances in the field of deep learning are expected to further reduce metal artifacts on medical images.
In summary, various approaches are employed to reduce metal artifacts on CT. Based on the findings described above, the combination of DECT and MAR is currently the best solution for metal artifact reduction. However, since MAR may create new artifacts in the case of lightweight metals, this combination may fail to effectively reduce metal artifacts [31]. Therefore, to adjust the metal artifact reduction approach, minimize artifacts, and optimize the image quality and diagnostic value of CT, it is important for clinicians to provide implant-specific information to radiologists because metal artifacts differ due to differences in the size, geometry, and alloys of metal implants [31] (Table 2).

Magnetic Resonance Imaging
Magnetic resonance imaging (MRI) plays an important role in the assessment of musculoskeletal disorders, including bone and soft tissues. However, due to heavy metal artifacts, MRI on a patient with metal implants leads to poor image quality and, thus, has been avoided [65]. Over the last few decades, various sequences have been introduced to obtain ideal images with fewer metal artifacts ( Figure 6). The metal artifact reduction sequence (MARS) was described by Olsen in 2000 [66]. Recent studies reported the usefulness of MARS for diagnosing periprosthetic infections of the shoulder [67] and hip and knee [68]. More advanced sequences for metal artifact reduction have been introduced, such as the view angle tilting (VAT) technique [69], WARP (a sequence commercialized by Siemens Healthcare, Erlangen, Germany) [65], slice encoding for metal artifact correction (SEMAC) [70], multi-acquisition variable-resonance image combination (MAVRIC) [71], and MAVRIC selective (MAVRIC SL) [72].
VAT is a strategy that is used to decrease in-plane artifacts. It reduces signal loss and pile-up artifacts by adding an altered readout gradient [69]. However, VAT causes a small degree of blurring and cannot correct through-section distortions [65]. WARP is intended to optimize the standard principles of MARS and includes multidirectional VAT to achieve further reductions in in-plane distortions [65]. SEMAC applies additional phase-encoding gradients perpendicular to the slice (z-direction) prior to signal readout in order to correct through-plane distortions, such as non-planar slice excitation [70,71]. It corrects metal artifacts by the strong encoding of each excited slice against metal-induced field inhomogeneities, which addresses both in-plane and through-plane distortions [71]. VAT and SEMAC were reported to be useful in evaluations of patients after THA and TKA [73,74].
Galley et al. examined 40 patients with periprosthetic infections after THA using 1.5 T MRI based on the coronal short inversion time inversion-recovery (STIR)-SEMAC sequence. The detection of a periosteal reaction, capsule edema, and intramuscular edema in evaluations of periprosthetic joint infections had sensitivities of 78, 83, and 95%, specificities of 90, 95, and 86%, and accuracies of 86, 91, and 89%, respectively. Interobserver agreement was almost perfect, with ICC values between 0.88 and 0.92 [75]. Takahashi et al. investigated the reliability of SEMAC to detect prosthesis loosening by comparing it to surgical outcomes in 47 patients after THA on 55 hips. Patients were divided into two groups, those with a painful hip (group P) and those without pain (group C). MRI was scored for osteolysis and bone marrow edema. Eleven patients in group P required revision surgery, in contrast to 0 in group C. Correlations between SEMAC and revision surgery outcomes were moderate to weak (Γ = 0.415 on T1W-SEMAC, and Γ = 0.35 on STIR-SEMAC). Sensitivity, specificity, positive predictive value, and negative predictive value (NPV) in group P were 72.7, 64.3, 44.4, and 85.7%, respectively, with T1W-SEMAC, 90.9, 46.4, 40.0, and 92.9%, respectively, with STIR-SEMAC, and 36.3, 78.5, 40.0, and 75.8%, respectively, with proton density-weighted (PDW)-SEMAC. They concluded that STIR-SEMAC achieved high sensitivity and NPV for detecting periprosthetic fluid and marrow edema, indicating prosthetic loosening [76]. VAT is a strategy that is used to decrease in-plane artifacts. It reduces signal pile-up artifacts by adding an altered readout gradient [69]. However, VAT cause degree of blurring and cannot correct through-section distortions [65]. WARP is i to optimize the standard principles of MARS and includes multidirectional achieve further reductions in in-plane distortions [65]. SEMAC applies additiona encoding gradients perpendicular to the slice (z-direction) prior to signal readout to correct through-plane distortions, such as non-planar slice excitation [70,71]. It metal artifacts by the strong encoding of each excited slice against metal-induc inhomogeneities, which addresses both in-plane and through-plane distortions [7 and SEMAC were reported to be useful in evaluations of patients after THA a Jungmann et al. compared the combination of VAT and SEMAC with conventional MRI to assess its usefulness in evaluations of large orthopedic tumor endoprosthesis in 25 patients. They used a five-point scale to assess qualitative parameters. Artifact diameters and distortions were significantly reduced in the VAT and SEMAC groups. Two cases of tumor recurrence, ten of infection, and thirteen other pathologies were diagnosed. They suggested that the combination of VAT and SEMAC was beneficial for detecting periprosthetic pathologies in postoperative follow-ups due to significant reductions in metal artifacts [69].
MAVRIC is a spin-echo-based sequence that minimizes image distortions by combining multiple individual datasets acquired at frequency bands that are gradually offset from the dominant proton frequency. It is capable of reducing both through-section and in-plane artifacts [65,71]. MAVRIC SL has the advantages of both STIR and MAVRIC. STIR allows for more uniform fat suppression around metals than spectral fat saturation. MAVRIC SL maintains metal artifact corrections and the high signal-to-noise ratio (SNR) of MAVRIC using an overlapped spectral strategy with multiple frequency-selective excitations [72]. Zochowski et al. conducted a feasibility study to compare the quality and diagnostic utility of three metal artifact reduction sequences in evaluations of 84 patients after THA. The investigated sequences were MAVRIC SL, isotropic MAVRIC SL, and reduced repetition time isotropic MAVRIC SL using a 1.5T imager. In comparisons with conventional MAVRIC SL acquisitions, they concluded that isotropic MAVRIC SL acquisitions improved SNR, the findings of lesions, such as the presence of low-signal-intensity deposits, osteolysis, the loosening of a prosthesis, and the visualization of synovium and periprosthetic bone, and reduced blurring. Interrater and intrarater agreement were substantial to almost perfect for the clinical features investigated (ICC values 0.61-1.00). However, overall SNR and visualization of synovium on conventional MAVRIC SL images tend to have a low agreement, with an ICC value of 0.26 and 0.08, respectively [77]. Kim et al. conducted a preliminary study to investigate the usefulness of MAVRIC SL STIR in evaluations of the postoperative cervical spine with artificial disk replacement among five volunteers and a cadaver using a 3T imager. The images obtained with MAVRIC SL STIR were compared with STIR images. MAVRIC SL STIR showed fewer signal void areas and distortions and better visualization of anatomical structures. However, the image quality of the spinal cord was better with STIR. For the cadaveric study, interobserver agreement was substantial (κ = 0.7). For the volunteer study, the agreement was almost perfect (κ = 0.89). They concluded that MAVRIC SL STIR may be useful in the evaluations of surgical sites after artificial disk replacement by signal void reductions and distortions [72].
As imaging sequences become more complex, imaging times have increased. Recent advances in acceleration techniques allow for more precise and rapid examinations. Parallel imaging (PI), simultaneous multi-slice acquisition, compressed sensing (CS)-based sampling, and synthetic MRI techniques are acceleration techniques that shorten acquisition times in a linear manner. CS acceleration is ideally suited for SEMAC and MAVRIC due to its high intrinsic sparsity [78].
In summary, more advanced imaging sequences for metal artifact reduction on MRI have been introduced in recent years. The combination of the CS acceleration technique with SEMAC currently appears to be the best solution in MRI to achieve metal artifact reductions with a reasonable acquisition time [87]. An MR imager with a stronger magnetic field, such as 7-Tesla, has recently been used, and the advanced metal artifact reduction techniques described herein appear to be efficient [88]. It is important to remain up-to-date on advances in imaging techniques and select the most suitable criteria for each patient (Table 3).

Future Perspectives and Conclusions
The number of patients undergoing arthroplasty surgeries has increased worldwide to more than one million per year [1]. In addition to arthroplasty surgeries, surgeons use intramedullary nails and locking plates in patients with various fractures. Increases in the number of surgeries are associated with a higher number of complications, such as periprosthetic infections and fracture non-union. Since their rapid diagnosis and treatment are favorable, metal artifact reduction techniques are essential for a precise diagnosis. Accurate imaging is necessary for the appropriate decision-making regarding treatments.
Advances have been achieved in diagnostic imaging technologies, and novel technologies are continuously being introduced. However, this review indicates that the latest technology has not yet been fully applied clinically. Clinicians must constantly update their knowledge and work closely with radiologists to select the best diagnostic imaging method for each case (Table 4). • Traditionally, heavy metal artifacts, such as signal loss and distortion, lead to poor image quality.

•
As imaging sequence becomes more complex, imaging time increases. Combined use of acceleration techniques is necessary to reduce imaging time.
As reviewed herein, there are various approaches to reducing metal artifacts. Different algorithms need to be selected based on the size, shape, material and implanted body parts of the implant. Future advances in metal artifact reduction algorithms and techniques and the development of new sequences will enable further reductions in metal artifacts even on original images taken previously. In addition, the combination of different imaging modalities may contribute to further developments in the field of metal artifact reductions. In intraarticular lesion imaging, a novel technique for the three-dimensional fusion of MRI and CT has been introduced [89,90]. Each imaging technique may compensate for the other's shortcomings and contribute to an accurate diagnosis by clinicians.