Keywords
MRI - CT - implant - hip arthroplasty - dual energy - metal artifact reduction
Improvements in implant design and surgical techniques, along with higher demands
for maintained mobility despite older age, have drastically increased the number of
patients undergoing hip replacement surgery. As one of the surgical procedures with
the best outcome, total hip arthroplasty (THA) is estimated to reach an annual rate
of 570,000 in the United States[1] and 324,000 in the United Kingdom[2] by 2030.
Complications associated with THA may originate from the implant itself, surrounding
osseous structures, periprosthetic soft tissues, or synovial reaction.[3] Although complication rates are low,[4] the high prevalence of the hip replacement procedures will result in an overall
growing number of patients with complications that will be encountered in daily radiology
practice.
Conventional radiography is the primary imaging tool for the routine surveillance
of patients following hip replacement surgery and to investigate symptomatic individuals.
Cross-sectional imaging is reserved for further characterization of radiographic abnormalities
or investigation of radiographically occult complications. Owing to recent developments
in imaging techniques, an accurate diagnosis of THA complications in their early stages
has now become feasible.
In this article, we review the basis of metal artifacts caused by THA implants on
magnetic resonance imaging (MRI) and computed tomography (CT) and explain basic and
advanced metal artifact reduction (MAR) techniques, as well as practical tips and
tricks to optimize CT and MR imaging of hip arthroplasty implants.
Metal-related MRI Artifacts
Metal-related MRI Artifacts
Image quality in the presence of metal is impaired by the inhomogeneity of the static
(B0) and radiofrequency (B1) magnetic fields. The B1 field inhomogeneity in the vicinity of metal is particularly accentuated by shielding
of the radiofrequency (RF) pulse and local electric fields induced by switching gradient
fields.[5]
[6]
[7] In comparison with B1, inhomogeneity of the B0 magnetic field has been studied more extensively and forms the basis for all MAR
techniques currently available for clinical use.[8] Perturbations in B0 in the vicinity of metal occur in various degrees depending on the hardware material
type, orientation, and configuration, and they result in three broad categories of
artifacts: spatial misregistration, signal loss, and failed fat suppression.
Spatial Misregistrations
To form an image in MRI, spatial localization of each voxel of tissue is achieved
by applying position-dependent gradient fields during slice selection and readout.[9] As such, the location of each spin ensemble is linearly related to the local magnetic
field in that particular location, and hence to the spin precession frequency. Metal-related
B0 inhomogeneities violate this linearity by altering the precession frequency of the
affected spins. As a result, spins outside the slice of interest are excited during
slice selection and wrongfully contribute to the formed image. Similarly, during readout,
pixels are misregistered to the wrong locations along the frequency-encoding (readout)
direction. Such misregistrations appear as geometric distortion, signal loss, or pileup[9]
[10] ([Fig. 1]). Because all lines of the k-space undergo a similar phase shift as a result of
metal-induced B0 perturbations, unlike slice- and frequency-encoding processes, phase encoding is
immune to spatial misregistrations.
Fig. 1 Axial intermediate-weighted turbo spin-echo MR image of a 66-year-old man with right
total hip replacement demonstrates geometric distortion, signal loss, and signal pileup
(arrows) that occur in a frequency-encoding direction. The phase-encoding direction
(right to left) is immune to such effects.
Signal Loss
Significant variation of the local magnetic field within a single voxel that can occur
in the vicinity of metallic implants leads to rapid dephasing and incoherence of the
intravoxel spins and appears as a dark area of signal void surrounding the implant
([Fig. 1]). Signal loss may also be due to failed excitation of those periprosthetic spins
that resonate at a frequency outside the bandwidth of the RF pulse.[8]
Failed Fat Suppression
Chemical-shift-selective fat suppression benefits from the different resonance frequency
of fat and water protons. By applying a saturation pulse tuned to the fat resonance
frequency, it selectively suppresses the fat signal. Metal-related B0 inhomogeneity shifts the fat peak outside the frequency-specific saturation pulse,
resulting in failure of fat suppression ([Fig. 2]). It may also suppress the water signal by shifting the water precession frequency
into that of the fat tuned suppression pulse.
Fig. 2 Coronal T2-weighted MR image with spectral fat suppression technique of a 73-year-old
man with left total hip replacement demonstrates the failure of fat suppression (arrow)
around the hip implants due to shifting of the periprosthetic fat spin frequencies
that consequently no longer fall within the frequency-specific saturation pulse.
Basic Strategies of MRI Metal Artifact Reduction
Basic Strategies of MRI Metal Artifact Reduction
Imaging at Lower Field Strength Magnets
Susceptibility-induced field inhomogeneity is linearly proportional to the field strength,
and therefore one expects less metal-related artifacts at 1.5 T and more metal-related
artifacts at 3T. However, clinical MAR techniques can be implemented on both 1.5 and
3T scanners. The higher signal-to-noise ratio (SNR) of 3T imaging offers flexibility
for higher image quality and together with advanced and carefully optimized MAR methods
make reasonable artifact reduction possible at 3T as well.
Spin-echo–based Pulse Sequences
Spin-echo–based pulse sequences effectively mitigate signal loss due to intravoxel
dephasing by applying a 180-degree pulse to refocus the dephasing spins ([Fig. 3]).[10]
Fig. 3 Axial intermediate-weighted MR images of a 56-year-old man with a right total hip
implant using high receiver bandwidth (700 Hz/pixel). (a) Turbo spin-echo and (b) gradient-echo techniques demonstrate superior performance of the spin-echo–based
techniques over the gradient-echo techniques in reduction of the metal-related artifacts
(arrows).
Selection of Phase and Frequency-Encoding Directions
Because phase encoding is unaffected by spatial misregistrations, the user may swap
the phase- and frequency-encoding directions with the aim of displacing the artifacts
in a direction that causes less tissue obscuration. This, however, may occur in time
penalties due to the new need for phase oversampling to overcome potential wrap artifacts.
High Receiver Bandwidth
Spin-echo class pulse sequences can be optimized for MAR by increasing the bandwidth
of the receiver that reduces the number of voxels across the signal displacement extends
([Fig. 4]). Increasing the receiver bandwidth additionally results in improved edge sharpness
but also lower SNRs.[11] [Table 1] shows the Johns Hopkins MRI protocol for patients with hip arthroplasty, composed
of high bandwidth turbo spin-echo pulse sequences.
Fig. 4 Schematic demonstration of the effect of receiver bandwidth on signal displacement
in the frequency-encoding dimension. Metal-induced magnetic field inhomogeneities
cause a shift of the precession frequency (Δf) of spins at a particular location. At low receiver bandwidths, this frequency shift
results in a signal displacement (Δx
LBW). The same frequency shift (Δf) results in smaller signal displacement at high receiver bandwidths (Δx
HBW). The same principle applies to excitation pulse bandwidth and displacements along
the slice-selection dimension. BW, bandwidth; FOVx = field of view in the frequency (x) direction; HBW, high bandwidth; LBW, low bandwidth.
Table 1
Conventional metal artifact reduction protocol for MR imaging of hip arthroplasty
implants
|
Parameters
|
Coronal IW TSE
|
Coronal STIR TSE
|
Sagittal IW TSE
|
Sagittal STIR TSE
|
Axial IW TSE
|
Axial STIR TSE
|
Axial T1 TSE[a]
|
|
TE/TR, ms
|
30/3,800
|
6.3/3,000
|
30/3,800
|
6.3/3,000
|
30/3,800
|
7/3,000
|
6.8/650
|
|
Receiver bandwidth, Hz/pixel
|
504
|
501
|
504
|
501
|
504
|
510
|
504
|
|
No. of slices/Flip angle, degrees
|
27/150
|
21/140
|
31/150
|
21/140
|
35/150
|
31/140
|
25/140
|
|
Field of view, mm2
|
270 × 270
|
300 × 300
|
270 × 270
|
300 × 300
|
230 × 230
|
230 × 230
|
300 × 300
|
|
Matrix
|
320 × 70%
|
256 × 80%
|
320 × 70%
|
256 × 80%
|
320 × 80%
|
256 × 80%
|
205 × 70%
|
|
Slice thickness/Gap, mm
|
3.5/0
|
4/0
|
3.5/0
|
4/0
|
4/0
|
4.5/0.4
|
5/0
|
|
No. of excitations/Concatenations
|
3/1
|
2/2
|
3/1
|
2/2
|
3/1
|
3/1
|
1/1
|
|
Acceleration factor
|
2
|
2
|
2
|
2
|
2
|
2
|
2
|
|
Acquisition time, min:s
|
3:31
|
3:44
|
3:31
|
3:56
|
2:45
|
2:56
|
2:45
|
Abbreviations: IW, intermediate weighted; MR, magnetic resonance; STIR, short tau
inversion recovery; TE, echo time; TR, repetition time; TSE, turbo spin echo.
a Optional use for pre- and postcontrast gadolinium-enhanced MRI.
High Radiofrequency Pulse Bandwidth
Similar to the receiver bandwidth, increasing the bandwidth of the excitation pulse
decreases the number of slices across which the signal misregistrations propagates.
This comes with a penalty of increased specific absorption rate (SAR) deposition.
Imaging with Thinner Slices
Assuming fixed RF pulse bandwidth, thinner slices are achieved by applying stronger
slice-selection encoding gradients. Sharing the same concept with high bandwidth RF
pulses, this results in less metal-related artifacts. Although there is no increase
in SAR, thinner slices are associated with decreasing SNR.
Increasing the Image Matrix Size
Increased matrix size results in smaller voxels associated with reduced intravoxel
signal loss and increased conspicuity of metal-related artifacts, both resulting in
improved image quality.[12] It has no or minimal effect on in-plane distortions if other scan parameters are
left unchanged. Similarly, increasing the echo train length has no direct effect on
MAR.[13] However, platform-specific associated effects due to changes to the receiver bandwidth
in the background may result in apparent MAR effects.
Using Short Tau Inversion Recovery and Dixon Methods for Fat Suppression
Short tau inversion recovery (STIR) is a reliable technique for achieving homogeneous
fat suppression of tissues surrounding metal implants. Acting based on the different
T1 values of water and fat, STIR is immune to metal-induced field perturbations.[14] In the presence of metal, residual failed fat suppression in STIR can be eliminated
by matching of the bandwidths of the inversion and excitation pulses.[15]
[16] Despite its superior fat suppression, STIR is of limited value in postcontrast fat
suppression because contrast-enhanced tissues may also be nulled due to their reduced
T1 relaxation times. Dixon-based techniques acquire in- and opposed-phase images and
allow for secondary water-only and fat-only image reconstructions. Compared with STIR,
the fat suppression ability of Dixon in the presence of metal is inferior.[17] However, it can facilitate successful postcontrast MRI. In the absence of patient
motion, postprocessing subtraction of the pre- and postcontrast T1-weighted images
with identical image parameters results in the most accurate fat suppression when
metal implants are present ([Fig. 5], [Table 1]).[18]
Fig. 5 Axial T1-weighted (a) precontrast and (b) postcontrast SEMAC turbo spin-echo (TSE) MR images of a 62-year-old patient with
right total hip replacement (arrows) demonstrate the improved visualization of areas
of contrast enhancement on the subtraction image (c, asterisk), created through the subtraction of the precontrast (a) from the postcontrast
(b) image.
Advanced Techniques of MRI Metal Artifact Reduction
Advanced Techniques of MRI Metal Artifact Reduction
View Angle Tilting
Metal-related signal displacement along the slice-selection and frequency-encoding
directions is proportional to the corresponding gradient field strengths in these
directions. View angle tilting (VAT) applies this principle to decrease in-plane misregistrations
by replaying the slice-selection gradient field during readout[19] ([Fig. 6]). This added gradient field tilts the readout direction, with the slope of the tilt
being the ratio of slice selection to frequency-encoding gradient fields, making it
parallel to the spatial signal displacements. Accordingly, in-plane signal distortions
are mitigated, albeit at the expense of degrees of blurring ([Fig. 7]).[20] From a practical standpoint, when adequate artifact reduction is achieved at increased
receiver bandwidths > 500 Hz/pixel, application of VAT may not result in additional
MAR given its image blurring effect ([Fig. 7]).[8] The combination of VAT and isotropic three-dimensional (3D) fast spin-echo sequences,
such as sampling perfection with application-optimized contrasts by using different
flip angle evolutions (SPACE; Siemens Healthineers, Erlangen, Germany), has effectively
reduced metal artifacts about a variety of implant types.[21]
[22] However, 3D imaging of a sizable anatomical region such as the hip joint remains
challenging due to time-consuming oversampling requirements in the two phase-encoding
directions.
Fig. 6 Schematic illustration of the view angle tilting (VAT) technique. (a) Two adjacent blocks of tissue are represented as solid gray and white. The solid
gray block is not affected by metal-related field distortions; the adjacent white
block is misregistered to a new position (cross-hatched). In-plane and through-plane
displacements along the readout (x) and slice-selection (z) directions are shown by
Δx and Δz, respectively. The ratio of displacement (tan (ϕ) = Δx/Δz) is equal to Gs/Gf,
where Gs and Gf represent slice selection and readout gradient fields, respectively.
(b) In conventional imaging (b), readout in x direction results in remarkable in-plane
signal displacement, shown as the overlap between the two blocks, whereas with the
use of VAT (c), replaying the slice-selection gradient during readout practically tilts the readout
direction at the similar angle of ϕ. The image acquired with this tilt results in
less apparent in-plane displacements, although it comes at the expense of the introduction
of blurring at the junction of the two blocks.
Fig. 7 Coronal intermediate-weighted turbo spin-echo MR image of a 78-year-old woman with
left total hip arthroplasty implants show the application of the view angle tilting
(VAT) technique for the reduction of metallic artifacts (arrows). Images a and b were
acquired with a low receiver bandwidth of 150 Hz/pixel, (a) without and (b) with the VAT technique. In (b), the metal artifacts (arrow) are reduced by the VAT
technique, at the expense of markedly increased image blurring. Images (c) and (d)
were acquired with a high receiver bandwidth of 600 Hz/pixel, (c) without and (d) with the VAT technique. With a high receiver bandwidth, the metal artifact–reducing
effect of the VAT technique is minimized (c and d, arrows); where VAT technique still
introduces blurring (d), it is counteracted by the high bandwidth. (e) Image was acquired with a bandwidth of 600 Hz/pixel, the VAT technique, and 17 SEMAC-encoding
steps, resulting in almost no metal artifacts (arrow). SEMAC and VAT technique are
synergistic and achieve optimal results in combination.
Slice Encoding Metal Artifact Correction
Slice Encoding for Metal Artifact Correction (SEMAC) is a two-dimensional (2D) imaging
method that adds multiple spatial partitions to the turbo spin-echo pulse sequence
to mitigate the metal-related through-plane artifacts.[23] Fundamentally, SEMAC exploits an additional phase-encoding step in the slice-selection
dimension to resolve the z-location of each distorted slice profile, similar to the
way the z-direction is encoded in 3D imaging. For each slice, images from each spatial
partition are reconstructed separately and then combined into a final composite image
through linear or quadratic summation ([Fig. 8]). Furthermore, SEMAC may implement VAT to decrease in-plane distortions. Compared
with the conventional techniques, SEMAC has proven more efficient at metal-related
artifact reduction ([Fig. 7]).[24]
[25] The optimal number of additional phase-encoding steps in SEMAC, known as SEMAC steps,
is a compromise between the degree of artifact reduction and longer acquisition times.[26] Vendors may commercialize SEMAC with different names, such as O-MAR XD (Philips,
Best, Netherlands) and Advanced WARP (Siemens). [Table 2] shows the Johns Hopkins MRI protocol for patients with hip arthroplasty, composed
of turbo spin-echo SEMAC pulse sequences.
Fig. 8 Schematic diagram of the acquisition and composition process of a SEMAC-based pulse
sequence with 11 encoding steps for metal artifact reduction of a left hip replacement.
The final image (composite image) is a sum-of-square composition of the center image
and the signals from the spatial bins that successively provide the displaced periprosthetic
signal. Depending on the implant alloys, a higher number of SEMAC-encoding steps are
needed to “collect” the displaced signals.
Table 2
SEMAC metal artifact reduction protocol for MR imaging of hip arthroplasty implants
|
Parameters
|
Coronal IW TSE
|
Coronal STIR TSE
|
Sagittal IW TSE
|
Sagittal STIR TSE
|
Axial IW TSE
|
Axial STIR TSE
|
Axial T1 TSE[a]
|
|
TE/TR, ms
|
32/2,800
|
6.8/3,180
|
32/3,000
|
6.8/3,180
|
32/3,360
|
6.8/4,660
|
6.8/650
|
|
No. of SEMAC steps
|
13
|
11
|
13
|
11
|
13
|
11
|
11
|
|
Receiver bandwidth, Hz/pixel
|
504
|
501
|
504
|
501
|
504
|
501
|
504
|
|
No. of slices/Flip angle, degrees
|
27/140
|
21/140
|
31/140
|
21/140
|
35/140
|
31/140
|
25/140
|
|
Field of view, mm2
|
270 × 270
|
300 × 300
|
270 × 270
|
300 × 300
|
270 × 270
|
300 × 300
|
300 × 300
|
|
Matrix
|
320 × 70%
|
256 × 80%
|
320 × 70%
|
256 × 80%
|
320 × 75%
|
256 × 80%
|
192 × 70%
|
|
Slice thickness/Gap, mm
|
3.5/0
|
4/0
|
3.5/0
|
4/0
|
4/0
|
4/0
|
5/0
|
|
No. of excitations/Concatenations
|
1/1
|
1/1
|
1/1
|
1/1
|
1/1
|
1/1
|
1/1
|
|
Turbo factor/Acceleration factor
|
11/3
|
9/3
|
11/3
|
9/3
|
11/3
|
9/3
|
4/3
|
|
Acquisition time, min:s
|
7:12
|
7:14
|
7:20
|
7:36
|
6:51
|
7:29
|
6:53
|
Abbreviations: IW, intermediate weighted; STIR, short tau inversion recovery; TE,
echo time; TR, repetition time; TSE, turbo spin echo.
a Optional use for pre- and postcontrast gadolinium-enhanced MRI.
Multi-Acquisition with Variable-Resonance Image Combination (MAVRIC)
Field inhomogeneity around the metal hardware causes spin precession at a wide range
of frequencies, of which only a small subset, also known as “on-resonant spins,” are
excited by the RF pulse.[27] Lack of excitation of the off-resonant spins results in periprosthetic signal void.
In Multi-Acquisition with Variable-Resonance Image Combination (MAVRIC), this problem
was addressed by splitting each excitation into multiple frequency bins (known as
spectral bins) with discrete offsets in central frequencies. For each frequency bin,
a sub-image is acquired with a similar offset in the readout frequency.[28] The resultant sub-images are then combined through a sum-of-squares or maximum-intensity
projection scheme.[29] General Electric (GE, Milwaukee, WI) platforms may implement the MAVRIC-SL (MAVRIC-Selective)
variant, a hybrid form of MAVRIC and SEMAC.[30] Lack of slice selectivity is a major drawback of MAVRIC, requiring inflexible and
time-consuming 3D imaging. Recently, a rapid and flexible 2D version of MAVRIC was
proposed that excites a limited slice and spectral region using gradient reversal
between excitation and refocusing pulses.[31]
Acquisition Time Considerations
Acquisition Time Considerations
Although multispectral (MAVRIC) and multispatial (SEMAC) imaging techniques have substantially
mitigated metal-related artifacts, this improved image quality is coupled with longer
scan durations, owing to additional spectral bins in MAVRIC and spatial partitions
in SEMAC. To bring the acquisition times below the clinically viable levels, various
acceleration algorithms such as partial Fourier encoding and parallel imaging were
implemented.[30]
[32] Further acceleration has been achieved recently using compressed sensing (CS) that
saves time by pseudo-random sampling of the k-space and retrieves the lost data through
iterative image reconstruction. This technique was combined successfully with both
MAVRIC[33] and SEMAC.[34]
[35]
[36]
[37] With preserved image quality, SEMAC has gained an eightfold acceleration by exploiting
all synergies between parallel imaging and CS, resulting in acquisition times that
are similar to those of turbo spin-echo pulse sequences[34]
[35]
[36]
[37] ([Fig. 9]). The highly efficient combination of CS and SEMAC is expected to become available
for clinical practice in the near future. The optimal choice of CS-SEMAC steps and
iteration parameters for visualization of periprosthetic soft tissues was determined
in a recent study.[26]
Fig. 9 Coronal intermediate-weighted MR images of a 62-year-old woman with left total hip
arthroplasty implants. Comparison of conventional SEMAC with (a) threefold parallel imaging acceleration and (b) compressed-sensing SEMAC with factor 8 acceleration with otherwise the same pulse
sequence parameters including 19 SEMAC-encoding steps shows similar image quality,
whereas the acquisition times are threefold different.
Metal-related CT Artifacts
Metal-related CT Artifacts
Metal artifacts ([Fig. 10]) are seen in different grades of severity due to the eclectic range of metals, shapes,
and sizes used for hip arthroplasty implants.[38] These artifacts limit the accuracy for assessment of the bones, bone–metal interface,
and soft tissue structures, thus may render the images unreliable in some cases. This
is confounded by the fundamental nature of CT imaging, where multiplanar reconstructions
are all extrapolated from single-image acquisitions – unlike independent multiplanar
acquisitions of MR imaging.
Fig. 10 Axial computed tomography image of the pelvis in a patient with bilateral total hip
arthroplasty implants demonstrates beam hardening and scatter artifacts projecting
as black (white arrow) and white (black arrow) streaks, respectively.
Contributors to metal artifacts include several factors[39] that may result in reconstruction errors near metal implants due to corrupted data,
ultimately resulting in an inaccurate representation of tissues and possibly of abnormalities.
These effects are broadly categorized into four main domains that overlap in their
underlying etiology and effects including beam hardening, scatter, quantum noise and
photon starvation, and edge effects.
Beam Hardening
Beam hardening artifacts appear as dark streaks between heavily attenuating structures,
secondary to abrupt X-ray beam inhomogeneities and polychromatic scatter caused by
metal implants. This effect can be explained by the high density and high atomic numbers
of metals used for hip replacements when compared with background bone and soft tissues.
As X-ray beams pass through metal implants, the photon flux decreases with fewer low-energy
photos and much more numerous high-energy photons, resulting in “hardening” of the
X-ray beam.[40]
[41]
Scatter
Increased scatter of these high-energy photon beams confound standard reconstruction
algorithms that ultimately results in incorrect registration. The scattered photons
add to the measured intensity and lead to an underestimation of the absorption and
thus to dark streaks in the image, where white streaks are caused by an overestimation
of the absorption.[42]
Quantum Noise and Photon Starvation
Quantum noise refers to the statistical uncertainty of low photon flux due to the
quantum nature of the photon counts. It manifests as random bright and dark streaks
particularly appearing along the direction of highest attenuation. Photon starvation
can be seen in high-density metals and in metals with a high atomic number. It leads
to low photon counts and thus to increased noise and missing projection data. The
background signal of the detector also adds to the noise level when no photons are
detected at all.[40]
[43]
Edge Effects
Edge effects are observed at sharp edges between high and low attenuating tissues.
This effect also contributes to nonlinear partial volume and misregistrations of data.[43]
Basic Strategies of CT Metal Artifact Reduction
Basic Strategies of CT Metal Artifact Reduction
Multiple basic steps may be applied to data acquisition, image reconstruction, and
image visualization to improve the quality of CT images around hip arthroplasty implants.[44]
Increasing the tube voltage and current are recognized standard techniques to reduce
the magnitude of metal artifacts by allowing for more penetration power and photons,
respectively, resulting in less scatter and absorption and minimizing heterogeneity
in the exiting photon beam. This targets scatter, noise, and photon starvation artifacts
at the expense of an increased radiation dose.[44]
Reducing the detector size reduces the total scatter registered, thus minimizing overall
scatter artifacts. Similarly, using narrow collimation can reduce the partial volume
effects and simultaneously minimize scatter-related artifacts.[45]
[46]
Patient positioning, for example with bilateral hip arthroplasty implants not completely
aligned parallel to each other, and, where applicable, gantry tilt, which acts along
similar principles as the VAT MRI acquisition technique, may also reduce artifacts
in specific scenarios. During acquisition, acquiring the largest number of projections
per rotation may help reduce the extent of aliasing and undersampling artifacts, and
increase the resolution of fine details and intricate structures.[47]
Use of model-based or iterative image reconstruction, using a soft reconstruction
kernel instead of a sharp kernel, and reconstruction of the images from the acquired
raw data with thicker slices reduce the visual conspicuity of metal artifacts.[44] Iterative reconstruction algorithms apply a larger quantity of acquired data and
include photon statistics in the reconstruction, analyzed by intelligent extrapolations.
This in theory results in minimized scatter and reduced edge effects by using dedicated
correction algorithms.
Visualization of tissues around the metal hardware may be effectively improved by
the use of an extended Hounsfield unit scale.[12]
Advanced Techniques of CT Metal Artifact Reduction
Advanced Techniques of CT Metal Artifact Reduction
Dual-energy CT with Monochromatic Extrapolation
Dual-energy computed tomography (DECT) imaging reconstructs images acquired with two
photon spectra at different tube voltages, for example, 90 and 140 kV. This is achieved
simultaneously either by having two separate tubes for each specified kV, fast kV-switching
of tube voltage, using beam-split filters, or operating dual-layer detectors. This
is followed by using virtual extrapolated monoenergetic analysis to better evaluate
different tissues adjacent to one another, minimize artifacts, and evaluate structures
close to implants, in addition to the actual implants at various extrapolated monochromatic
energies.[48]
Virtual monochromatic imaging reduces beam hardening artifacts. Optimal monochromatic
energies vary for different kinds of metal hardware and tissue type, but most range
between 90 and 190 keV ([Figs. 11] and [12]). DECT, therefore, enables the reader to evaluate each tissue type at its optimal
energy level, without the need for additional imaging.[49]
[50]
Fig. 11 Computed tomography (CT) examination following greater trochanteric osteotomy and
slipped capital femoral epiphysis screw fixation. The visualized effects of metal
artifacts are often reduced when evaluating bone window images with (a) smooth kernel image reconstruction rather than (b) sharp kernel image reconstruction. (c) Sagittal virtual monoenergetic reconstruction at 190 keV shows the value of metal
artifact reduction using dual-energy CT monoenergetic evaluation for the assessment
of screw integrity, location of the screws in relation to the articular surface, extension
into the joint space, and contour of the femoral head. (d) Metal artifacts may also be reduced by applying a volume rendering technique to
evaluate the integrity of metallic implants.
Fig. 12 Periprosthetic osteolysis of left femoral hip arthroplasty component. (a) Coronal monoenergetic computed tomography (CT) image of the left femur at 180 keV,
(b) volume rendering CT image of the left femur, and (c) sharp X-ray-type volume rendering CT image demonstrate periprosthetic osteolysis
of the femoral component (arrow), varus deformity, and cortical erosion.
DECT artifact reduction proficiency is inversely proportional to increasing molecular
weight of the metal, larger volume implants, and metal implants with sharp edges.
Switching to higher beam energies might provide some additional artifact reduction;
however, in these cases, MAR software is often helpful.[51]
Metal Artifact Reduction Software
MAR algorithms are based on “projection completion,” also known as sinogram inpainting
techniques. Simply put, in this class of techniques, the corrupt X-ray projections
that traversed the metal hardware are removed and replaced with interpolation from
adjacent unaffected projections ([Fig. 13]). Some implementations may also benefit from iterative or model-based reconstructions
that additionally incorporate a more realistic model for image acquisition.[52] Several MAR algorithms were developed in the past. Prototypes include Normalized
MAR (NMAR),[53] Frequency Splitting MAR (FSMAR),[54] more recently iMAR (iterative metal artifact algorithms that combine the effects
of NMAR and FSMAR), adaptive mixing,[55]
[56] tissue modeling and adaptive filtering,[57] and iterative frequency splitting.[58]
Fig. 13 Total right hip arthroplasty implant in a 71-year-old man. (a) Comparison of conventional filter back-projection computed tomography (CT) image
reconstruction and (b) iterative metal artifact reduction CT image reconstruction with inpainting technique
demonstrates markedly reduced metal artifacts (arrows) in iterative metal artifact
reduction.
MAR algorithms, despite improving image quality, may obscure parts of the metal hardware,
alter the data of areas next to the metal edge, and introduce new artifacts.
Vendors provide MAR software under different commercial names including SEMAR (single-energy
MAR, Canon Medical Systems, Otawara, Japan), O-MAR (orthopaedic MAR, Philips Healthcare,
Best, Netherlands), SMAR and MARS (Smart MAR and MAR Sequence, respectively, GE Healthcare,
Milwaukee, WI), and MARIS and iMAR (MAR in Image Space and iterative MAR, respectively,
Siemens Healthineers, Erlangen, Germany). Specific details of these proprietary MAR
software are undisclosed. If not already included, the addition of iterative or model-based
reconstruction further increases the artifact reduction ability of such software.[59]
Combined MAR Software and Virtual Monochromatic DECT
The combination of MAR software and virtual monochromatic DECT imaging was advocated
in some cases because it may reduce metal-related artifacts more effectively than
one technique alone.[60]
[61]
[62] Limitations to combined use include affecting the appearance of metal implants,
over- or underestimating the size of the implant, and maybe even introducing secondary
artifacts.[61]
[63]
[64] Depending on the implant composition, size, and shape, no beneficial effects may
arise from combining both virtual monochromatic DECT imaging and iterative MAR.[65]
Three-dimensional Postprocessing
In volume rendering, data integration and averaging from consecutive axial planes
into the reformation planes leads to weighing of the true signal over the randomly
distributed artifacts that may result in improved image quality and often a visible
reduction of metal artifacts ([Fig. 14]).[66]
[67]
[68]
Fig. 14 Total right hip arthroplasty implant in a 78-year-old man with eccentric polyethylene
liner wear. Comparison of (a) coronal multiplanar reformation and (b) coronal volume rendering technique computed tomography (CT) images, both created
from the same conventional filter back-projection CT data set. Volume rendering technique
(b) can result in a visible reduction of low-density metal artifacts (arrows) when
compared with conventional multiplanar reformation (a).
When compared to conventional volume rendering postprocessing techniques,[69] cinematic rendering is a recently introduced 3D visualization method of volumetric
CT and MRI data that applies a highly sophisticated lighting model, enabling the generation
of photorealistic images resembling gross anatomical specimens ([Fig. 15]).[70] Sharing a similar concept with volume rendering, cinematic rendering may also reduce
tissue obscuration by metal-related artifacts.[52]
Fig. 15 Total left hip arthroplasty implant in a 59-year-old woman with heterotopic ossification.
(a) Axial and (b) sagittal oblique multiplanar reformation computed tomography images demonstrate
mature osseous bridging (arrows) between the femur and acetabulum consistent with
Brooker class 4 (bone ankylosis), the magnitude of which is visualized to better advantage
with cinematic rendering technique (c).
Radiation Dose Considerations
Smart machine designs allow for automated and more accurate estimations of the required
X-ray energies and tube currents for optimized CT imaging, based on an individualized
patient approach. This helps minimize radiation exposure by ensuring that patients
receive the smallest possible dose tolerable for a good quality diagnostic study.[71]
Evolving algorithms allow lower dose CT acquisitions. Utilizing deep learning artificial
intelligence methods. CT imaging of THA phantoms using iterative MAR software resulted
in the ability to maintain quantitative image quality parameters while reducing CT
radiation dose up to 80%.[72] Furthermore, implementation of an iterative MAR software did not compromise the
accuracy of lesion detectability near hardware while reducing CT radiation dose by
50%.[73]
Carefully constructed DECT protocols ([Table 3]) can have comparable effective radiation dose ranges when compared with single-energy
CT protocols.
Table 3
Single and dual-energy protocol for CT imaging of hip arthroplasty implants
|
Technical parameter
|
Value: Single energy
|
Value: Dual energy
|
|
Tube energy, kVp
|
120
|
100/Tin (Sn) filtered 150
|
|
Reference tube current with dose modulation, mAs
|
147
|
300/Adapted to tube A dose modulation
|
|
Rotation time, s
|
0.5
|
0.5
|
|
Collimation
|
192 × 0.6 mm
|
2 × 192 × 0.6 mm
|
|
Pitch
|
0.8
|
0.5
|
|
Average acquisition time, s
|
3
|
6
|
Conclusion
Rapid advances in metal implant designs and MAR techniques in CT and MRI, as well
as evolving patient selection criteria, have significantly improved the visualization
and diagnosis of hip arthroplasty implant-related abnormalities that not long ago
would have been otherwise undetectable. In clinical practice, an imaging protocol
composed of optimized conventional and advanced MAR pulse sequences enables substantial
artifact reduction in clinically reasonable acquisition times. DECT with virtual monoenergetic
reconstruction, MAR algorithms, or their conjoint implementation can remarkably reduce
metal artifacts and improve diagnostic image quality in most cases. Due to the different
nature of post-acquisition data analysis and interpolations, CT images reconstructed
using MAR algorithms should be interpreted cautiously and in line with the original
data sets. New innovations in optimizing imaging aim at refining the algorithms for
both CT and MR to further reduce artifacts and adapt to new metal implant designs
while aspiring for reductions in acquisition times for MRI and radiation dose for
CT.
