PMCCPMCCPMCC

Search tips
Search criteria 

Advanced

 
Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
 
Magn Reson Med. Author manuscript; available in PMC 2010 July 1.
Published in final edited form as:
PMCID: PMC2837371
NIHMSID: NIHMS171941

SEMAC: Slice Encoding for Metal Artifact Correction in MRI

Abstract

Magnetic resonance imaging (MRI) near metallic implants remains an unmet need due to severe artifacts, which mainly stem from large metal-induced field inhomogeneities. This work addresses MRI near metallic implants with an innovative imaging technique called “Slice Encoding for Metal Artifact Correction” (SEMAC). The SEMAC technique corrects metal artifacts via robust encoding of each excited slice against metal-induced field inhomogeneities. The robust slice encoding is achieved by extending a view-angle-tilting (VAT) spin-echo (SE) sequence with additional z-phase encoding. While the VAT compensation gradient suppresses most in-plane distortions, the z-phase encoding fully resolves distorted excitation profiles that cause through-plane distortions. By positioning all spins in a region-of-interest to their actual spatial locations, the through-plane distortions can be corrected by summing up the resolved spins in each voxel. The SEMAC technique does not require additional hardware and can be deployed to the large installed base of whole-body MRI systems. The efficacy of the SEMAC technique in eliminating metal-induced distortions with feasible scan times is validated in phantom and in vivo spine and knee studies.

Keywords: distortion correction, metallic implants, metal artifacts, metal-induced distortions, susceptibility artifact, field inhomogeneities, distorted excitation profiles, view-angle-tilting

Introduction

Metallic implants, such as pedicle screws, are commonly used in orthopedic surgery to fixate fractures, replace arthritic joints, and to align and immobilize vertebra. In the United States alone, there were 325,000 spinal fusions performed in 2003 and 450,000 primary or revision total knee arthroplasties performed in 2002 [1]. The current standard imaging test for complications associated with metallic implants is a plain radiography. For accurate diagnosis, a radiography requires the x-ray beam to be oriented exactly parallel to the bone-implant interface: any obliquity of the x-ray beam can obscure the radiolucent area [2]. A better substitute for a two-dimensional radiograph is cross-sectional imaging that images the entire bone-implant interface in three dimensions. However, the physical characteristics of metallic implants cause difficulties with cross-sectional imaging techniques. Computed tomography (CT) and invasive CT myelography suffer from metal-induced streak/beam-hardening artifacts and data loss throughout the field-of-view (FOV) [3]. In addition, radiography and CT are relatively insensitive to soft tissue abnormalities and bone marrow edema caused by infection, which are important complications to be evaluated in patients with implants.

Magnetic resonance imaging (MRI) is potentially the best imaging modality for diagnosing patients with metallic implants, due to its superior soft tissue contrast [4]. However, MRI near metallic implants is hampered by severe artifacts, which stem from large metal-induced field inhomogeneities [5], local gradient-induced eddy currents on metal surfaces [6], and radiofrequency (RF) shielding effects [7]. Among these, metal-induced field inhomogeneities are responsible for most severe artifacts. It is well understood that steep field gradients near metal objects result in increased intra-voxel dephasing, and a severely shortened equation M1. As a result, MRI near metallic implants inevitably involves spin-echo (SE) sequences, which refocus the dephased spins. However, spin-echo techniques still suffer from spatially dependent artifacts (e.g., signal voids and pile-ups), owing to a non-linear frequency-position mapping caused by metal-induced field inhomogeneities. In the absence of field inhomogeneities, the mapping between a spin's precession frequency and its spatial location is a linear function (dotted line in Fig. 1), and the slope of the linear mapping is determined by the gradient amplitude. When metal-induced field inhomogeneities (dashed line of a bell shape in Fig. 1) superimpose upon the frequency induced by the gradient, the resulting frequency-position mapping becomes highly non-linear (solid line in Fig. 1), which causes problems in slice-selective excitation and frequency encoding during the readout.

1
In the absence of field inhomogeneities, the mapping between a spin's precession frequency and its spatial location is a linear function (dotted line), and the slope of the linear mapping is determined by the gradient amplitude. When metal-induced field ...

As illustrated in Fig. 1, an RF pulse with 1 kHz bandwidth centered at 11.5 kHz is designed to excite a 3 mm-thick slice at the slice location 3.3 cm. However, due to the non-linear frequency-position mapping, the RF pulse excites a much thinner slice at the slice location 1.8 cm, resulting in through-plane displacement and signal loss due to the thinner slice. In a more extreme case, an RF pulse with 1 kHz bandwidth centered at 14.5 kHz excites spins centered around three locations (light gray regions around 2.3, 3.1 and 4.4 cm), resulting in through-plane pile-ups. This non-linear frequency-position mapping can also be used to demonstrate the formation of in-plane distortions, when the gradient acts as a readout gradient. For example, the spins from the three light gray regions (around 2.3, 3.1, and 4.4 cm) are subject to the same frequency encoding, which leads to in-plane pile-ups along the readout direction. Based on the directions along which spatially dependent artifacts exist, the artifacts are grouped into through-plane distortions and in-plane distortions. These two types of distortions occur at different stages in a pulse sequence: through-plane distortions result from a distorted excitation profile during the slice-selective excitation, while in-plane distortions are due to disrupted frequency encoding during the readout.

Correction of metal-induced artifacts through postprocessing often relies on some assumptions about metal objects and field maps to unwrap the distortions. As it is difficult to have reliable estimation of field inhomogeneities around metal, these works only show success in obtaining in vivo MR images near tiny metallic objects [8] and may encounter difficulty in resolving pile-ups [9]. In light of the distortions scaled with field strength, a prepolarized MRI (PMRI) system, which allows for low-field signal acquisition with mid-field signal-to-noise ratio (SNR), has been developed to image near metallic implants [10]. However, the techniques developed for PMRI cannot be generalized to conventional whole-body systems (e.g., 1.5 T scanners). Currently, the most successful works on MRI near metallic implants are based on the view-angle-tilting (VAT) technique developed by Cho et al. [11]. The work by Butts et al. [12] provides solutions to reduce VAT-associated blurring. In another work by Kolind et al. [13], a VAT-SE sequence with high RF and readout bandwidths is referred to as the “metal artifact reduction sequence” (MARS), which produces promising MR images near metal objects. However, these advances in employing VAT for imaging near metal objects only address the correction of in-plane distortions. Except the work by Butts et al. [9] that resamples slice profiles based on field maps, the work by Koch et al. [14] that performs multiple frequency offset acquisitions, and our preliminary work [15], most methods suffer from through-plane distortions.

In this work we address both in-plane and through-plane distortions with an innovative imaging technique called “Slice Encoding for Metal Artifact Correction” (SEMAC). The SEMAC technique corrects metal artifacts via robust encoding of each excited slice against metal-induced field inhomogeneities. The robust encoding is achieved by extending a VAT-SE sequence with additional phase encoding along the slice-select z-axis [15], as shown in Fig. 2. While the VAT-compensation gradient suppresses most in-plane distortions, the additional z-phase encoding is included to fully resolve a complicated distorted excitation profile of each slice. By positioning all spins in a region-of-interest to their actual spatial locations, the through-plane distortions can be corrected by summing up the resolved spins in each voxel. To avoid SNR degradation of the results obtained from the SEMAC technique, the relative phases between slices must be eliminated such that the spins resolved from different slices can be coherently summed without taking the magnitude operation. The SEMAC technique does not require additional hardware and can be deployed to the large installed base of whole-body MRI systems. The efficacy of the SEMAC technique in eliminating metal-induced distortions with feasible scan times is validated in phantom and in vivo spine and knee studies.

2
Schematic diagram of the SEMAC sequence for imaging near metallic implants. The VAT compensation gradient suppresses in-plane distortions, while the additional z-phase encoding steps are included to resolve distorted excitation profiles that cause through-plane ...

Methods and Materials

Let m(x, y, z) and Δf(x, y, z) denote spin density and field inhomogeneity at a voxel location (x, y, z), where x and z are assumed to be readout and slice-select directions, respectively. As there is no distortion along the phase-encoding y-axis, we drop the dependence of m(x, y, z) and Δf(x, y, z) on y for notational simplicity in the remainder of this section.

Distorted Excitation Profile

We first consider the effect of the field inhomogeneity Δf(x, z) during slice-selective excitation. Given an RF pulse of bandwidth BRF and a slice-select gradient of amplitude Gs, the excited magnetization me is given by:

equation M2
(1)

where z0 is the nominal slice location excited at a transmit frequency equation M3, and equation M4 is the gyromagnetic ratio equal to 42.58 MHz/Tesla for 1H. The through-plane shift Δz for a spin with field inhomogeneity Δf is

equation M5
(2)

where s = 2πBRF/(γGs) is the nominal slice thickness. [n-ary square intersection operator] (f/BRF) is the frequency profile of the RF pulse, which ideally is a rectangular function defined by

equation M6

The metal-induced field inhomogeneities cause the excited magnetization me(x, z) to contain spins from different slice locations, for which |z − Δz(x, z) − z0| ≤ s/2. Therefore, Δz(x, z) represents the distorted excitation profile caused by the field inhomogeneity Δf(x, z).

Figure 3 illustrates that distorted excitation profiles can lead to severe through-plane distortions. Figure 3 (a) depicts a simulated dipole pattern of metal-induced field inhomogeneities [16], where the sample field inhomogeneities along the x-axis (solid line) and along the z-axis (dashed line) are plotted in Fig. 3 (b). When the field inhomogeneities superimpose upon the frequency induced by the slice-select gradient, the resulting excitation profiles correspond to the frequency bands shown in Fig. 3 (c). The spins are excited when their precession frequencies fall into the corresponding frequency bands, and the excitation profiles of three sample slices are shown in Figs. 3 (d, e, f). As the excited slices contain spins from different slice locations, the distorted excitation profiles lead to through-plane distortions.

3
Illustration of severe through-plane distortions caused by distorted excitation profiles. (a) depicts a simulated dipole pattern of metal-induced field inhomogeneities from a cylindrical object, where the sample field inhomogeneities along the x-axis ...

SEMAC Sequence

Figure 2 shows the proposed pulse sequence, in which the z-phase encoding steps are included to resolve the distorted excitation profile of each slice. Given Nzz-phase encoding steps with an incremental amplitude of Gzi and a duration of Tz, the resolution of z-phase encoding is δs = 2π/(γNzGziTz). Besides the z-phase encoding, we also apply a VAT-compensation gradient of the amplitude Gs during the readout [11]. At the readout time t and the nthz-phase encoding step, the received signal r(t, n) is given by:

equation M7

Let us examine the effects of the VAT-compensation gradient during the readout:

equation M8
(3)

where δz(x, z) = zz0 − Δz(x, z) is the ambiguity in determining the slice locations of the spins in the distorted slice and |δz| ≤ s/2 (see Eq. 1). Equation 3 shows that δz is responsible for uncorrected in-plane spin distortions equation M9, as the encoded x-location is given by

equation M10
(4)

When expressed in terms of pixels, the uncorrected in-plane spin distortion is bounded by

equation M11
(5)

where Bp = 2Bread/Nx is the readout bandwidth per pixel. (Bread and Nx are the readout bandwidth and the number of readout samples, respectively.) In a typical imaging case, 1 kHz RF bandwidth, ±125 kHz readout bandwidth and 256 readout samples limit the maximum in-plane spin distortion to be less than 0.512 pixels with the VAT compensation gradient. With referring to the through-plane distortion Δz (Eq. 2) and the maximum in-plane spin distortion δx with the VAT technique (Eq. 5), it can be seen that the selection of the RF bandwidth BRF forces a tradeoff between Δz and δx (cf. Discussion).

Equation 3 shows that the VAT-compensation gradient also modulates the excited spins at the transmit frequency fe during the readout. In case of exciting multiple slices, we need to demodulate each slice with its transmit frequency accordingly such that the slices do not have relative in-plane shifts. The received signal r(t, n) demodulated at fe is given by

equation M12
(6)

where kzi = γGziTz is the z-phase encoding increment. The z-phase encoding FOVz is given by 2π/kzi, which should be sufficiently large to avoid aliasing in the resolved excitation profile.

Correction of Through-plane Distortions

Equation 6 shows that the proposed SEMAC sequence enables robust spatial encoding against metal-induced field inhomogeneities. An inverse Fourier transform of the received signal r(t, n) completely resolves the spatial distribution of the magnetization me(x, z). When distorted excitation profiles result in through-plane distortions, pile-ups in one slice correspond to signal voids in other slices. Hence, multiple slices are required in the correction of through-plane distortions. Figure 4 illustrates the working principle of the SEMAC technique. During the image acquisition, the metal-induced field inhomogeneities result in complicated distorted excitation profiles (highlighted in different colors). The number of the slices, NS, is chosen to cover all spins inside a region-of-interest (ROI). For each slice, its excitation profile is resolved with the z-phase encoding. During the image reconstruction, the spins in the ROI are positioned back to their actual voxel locations. Subsequently, the through-plane distortions can be corrected by summing up the spins in each voxel.

4
Illustration of the correction of through-plane distortions in the SEMAC technique. During the image acquisition, the metal-induced field inhomogeneities result in complicated distorted slice excitation profiles (highlighted in different colors). For ...

In practice, direct summation usually results in severe signal loss, as the spins resolved from different slices can have relative phases. The relative phases between slices can be understood by examining Eq. 3, where the VAT-compensation gradient modulates the spin echo in each slice to the corresponding RF transmit frequency. As the spins excited in different slices precess at different RF transmit frequencies, any timing mismatch in echo acquisition results in relative phases between the slices. While summing the magnitudes of the resolved spins removes the relative phases between slices, the magnitude operation causes the background noise to become Rayleigh-distributed, which is no longer zero-mean [17]. Summing the non-zero mean noise degrades the resultant signal-to-noise ratio (SNR). In this work, we eliminate the timing mismatch in echo acquisition by carefully adjusting the RF phase reference and/or the receive phase reference [18]. When no relative phase exists between the slices, the resolved spins in each voxel can be coherently summed without taking the magnitude operation.

SNR Analysis

To evaluate the SNR performance of the SEMAC technique, we consider an impulse object located at the origin (i.e., x = 0 and z = 0). To simplify the analysis, we assume that no field inhomogeneity disrupts slice-selective excitation and frequency encoding; hence, all SNR evaluations are conducted at the pixel that corresponds to the voxel centered at the origin. A one-dimensioned reconstruction of the object is taken as the reference case for SNR evaluation. The impulse object gives rise to a constant signal amplitude M for each of the Nx readout samples. Without taking the magnitude operation, each readout sample contains zero-mean white Gaussian noise with variance equation M13. The SNR of the 1D reconstruction is given by

equation M14

which is the SNR of the reconstructed impulse object obtained from a standard spin-echo sequence or a VAT-SE sequence.

The SEMAC sequence imposes the readout samples obtained from a VAT-SE sequence with z-dependent phase factors. As the impulse object resides at the origin, the slice location is z = 0; hence, all z-dependent phase factors are equal to 1. Therefore, for imaging this impulse object, the SEMAC sequence essentially repeats the VAT-SE sequence for Nz times. (Nz is the number of z-phase encoding steps.) The reconstruction of the impulse object is achieved with an inverse 2D Fourier transform evaluating at the origin, and the SNR with the SEMAC sequence is given by

equation M15

which shows that the SNR is improved by equation M16, given the longer scan time.

The gain in SNR with the SEMAC sequence is, however, reduced by the correction of through-plane distortions that involves summing multiple slices. Each slice has a z-phase encoding FOV (FOVz = Nzδs), which is centered at its nominal slice location. The relative shift along the slice-select z-axis is the nominal slice thickness between two neighboring slices. In our implementation, the z-phase encoding resolution is set to be the same as the nominal slice thickness; hence, the voxel corresponding to the origin can be covered by as many as Nz slices. While the impulse object only gives rise to non-zero signal amplitude for the slice that is centered at the origin, all k-space samples contain independent noise with variance equation M17. Consequently, the correction of through-plane distortion, which sums the resolved signal from different slices, results in an increase in noise variance. The resultant SNR is given by

equation M18

which is the same as the reference SNR obtained from the SE or VAT-SE sequence.

The above analysis shows that the SNR performance of the SEMAC technique is adversely affected by the number of slices involved in the correction of through-plane distortions. Therefore, the correction of through-plane distortions at a voxel should only involve the slices that contain the signals arising from this voxel. To that end, a heuristic threshold is selected to differentiate the signals from background noise. As the signals are generally much stronger than background noise and the frequency dispersion in each voxel is limited, the threshold is the median magnitude of all spins resolved from multiple slices. Only the spins with magnitudes greater than the threshold are included in the correction of through-plane distortions. In our experience, this threshold achieves a good balance between the correction of through-plane distortions and the exclusion of background noise.

Experimental Methods

All experiments were performed on a GE Signa 1.5 T scanner (General Electric Healthcare, Waukesha, WI) with gradients capable of 40 mT/m amplitude and 150 T/m/sec slew rate, and a readout bandwidth up to ±250 kHz. Reconstruction and post-processing were performed off-line in MATLAB (Mathworks, South Natick, MA). Windowed-SINC pulses with pulse duration 3.2 ms were used as both the 90° excitation pulse and the 180° refocusing pulse in the SEMAC sequence. The amplitude of the 180° refocusing pulse is twice that of the 90° excitation pulse. To alleviate the VAT-associated blurring, the readout bandwidths were chosen such that the readout duration matched with the main lobe of the RF pulses [12]; however, the relatively high readout bandwidths (i.e., greater than 100 kHz) lower the SNR of the resultant images. The number of slices and the nominal slice thickness were chosen to cover all spins in the region-of-interest. Multiple slices were excited and acquired in an interleaved manner. In the following experiments, 16 z-phase encoding steps were prescribed for each slice to resolve the slice excitation profile. For evaluation purposes, each of the following experiments was repeated with a spin-echo (SE) sequence and a VAT-SE sequence (by turning off the z-phase encoding steps), while keeping all other scan parameters the same as the SEMAC sequence.

Phantom Experiments

For phantom evaluation, the SEMAC technique was used to image a titanium shoulder prosthesis immersed in agar gel using a quadrature head coil and the following scan parameters: TE/TR = 11/400 msec, acquisition matrix = 256 × 128, FOV = 20 cm, readout bandwidth ±125 kHz, nominal slice thickness 2 mm with no gap between slices. The time-bandwidth product (TBW) of the RF pulse is 4, which gives the RF bandwidth BRF = 1.25 kHz. There are 16 z-phase encoding steps per slice, which account for ±10 kHz (i.e., ±8BRF) field inhomogeneities in each slice. To cover a volume containing the metallic implants, 22 slices were prescribed and 14-minute scan time was incurred.

In Vivo Experiments

For in vivo experiments, the SEMAC technique was used to image two subjects with metallic implants in their spines and one subject with several stainless steel screws in his lower leg. For imaging of the first subject with scoliosis rods in her spine, an eight-channel phased-array Cervical-Thoracic-Lumbar (CTL) coil and the following scan parameters were used: TE/TR = 11/523 msec, acquisition matrix = 256 × 128, FOV = 28 cm, readout bandwidth ±166 kHz, nominal slice thickness 3 mm with no gap between slices. The TBW of the RF pulse is 6.4, which gives the RF bandwidth BRF = 2 kHz. There are 16 z-phase encoding steps per slice, which account for up to ±16 kHz (i.e., ±8BRF) field inhomogeneities in each slice. To cover the spine and the implants, 32 slices were prescribed, resulting in an 18-minute scan time.

For imaging of the second subject with a metallic fixation device in his spine, an eight-channel phased-array CTL coil and the following scan parameters were used: TE/TR = 11/400 msec, acquisition matrix 256 × 128, FOV = 24 cm, readout bandwidth ±125 kHz, nominal slice thickness 4 mm with no gap between slices. The TBW of the RF pulse is 4, which gives the RF bandwidth BRF = 1.25 kHz. There are 16 z-phase encoding steps per slice, which account for up to ±10 kHz field inhomogeneities in each slice. To cover the spine and the device, 22 slices were prescribed. This study took a 14-minute scan time.

The last leg study used an eight-channel transmit/receive knee coil and the following scan parameters: TE/TR = 12/555 msec, acquisition matrix = 256 × 128, FOV = 32 cm, readout bandwidth ±166 kHz, nominal slice thickness 4 mm with no gap between slices. The TBW of the RF pulse is 6.4, which gives the RF bandwidth BRF = 2 kHz. There are 16 z-phase encoding steps per slice, which account for up to ±16 kHz field inhomogeneities in each slice. To cover a volume containing the knee, 32 slices were prescribed and 18-minute scan time was incurred.

Results

Figure 5 shows the phantom results obtained from the SE sequence, the VAT-SE sequence, and the SEMAC technique. As compared to the photo of the prosthesis, the SEMAC technique successfully recovers the shape of the prosthesis by suppressing severe distortions present in the results obtained from the SE sequence and the VAT-SE sequence. The first three rows show the comparison of the in-plane results. Despite partial correction of the distortions in the results obtained with the SE sequence, the VAT-SE sequence still suffers from uncorrected through-plane distortions, which occlude the features of the cobalt-chrome head of the prosthesis. In contrast, the SEMAC technique correctly reproduced the shape and fine details of the prosthesis. The improvement produced by the SEMAC technique can also be clearly visualized by comparing the reformatted results in the bottom row.

5
Comparison of the phantom results obtained from the SE sequence, the VAT-SE sequence, and the SEMAC technique. The first three rows show the comparison of the in-plane results. The dashed boxes in the first row select an uniform region for SNR evaluation ...

This phantom study is also used to evaluate the SNR performance of the SEMAC technique as compared to those of the SE and VAT-SE sequences. The SNR evaluation is conducted by selecting an uniform region in the in-plane results of a sample slice (identified with dashed boxes in Fig. 5). The SNR of each technique is empirically computed as the ratio of the average signal magnitude in the selected region to the standard deviation of background noise. The empirical SNRs of the SE sequence, the VAT sequence, and the SEMAC technique are 22, 22, and 34, respectively. Hence, by excluding background noise from the correction of through-plane distortions, we improve the SNR performance of the SEMAC technique.

Figure 6 shows the in vivo results of the subject with scoliosis rods in her spine. As compared to the SE (the first row) and VAT-SE (the second row) sequences, the SEMAC technique (the bottom row) achieved distortion-free imaging of the scoliosis rods. As noted by arrows in the first and third columns, the SEMAC technique managed to produce perfect delineation of the rods and pedicle hooks, which is important for determining if there is any displacement of the implants. In the second column, the SEMAC technique recovered the spinal cord, otherwise occluded in the results obtained from the SE and VAT-SE sequences, which is critical for evaluating spinal complications associated with the implants. The rightmost column in Fig. 6 shows the sample coronal reformats of the results, which demonstrate the efficacy of the SEMAC technique in correcting the metal-induced distortions hampering the SE and VAT-SE sequences.

6
Comparison of in vivo results of the subject with scoliosis rods in her spine. The results were obtained from the SE sequence, the VAT-SE sequence, and the SEMAC technique. In comparison with the SE (the first row) and VAT-SE (the second row) sequences, ...

Figure 7 shows the in vivo results of the subject with a metallic fixation device in his spine. The results were obtained from the SE sequence (top row), the VAT-SE sequence (middle row) and the SEMAC technique (bottom row). The SEMAC technique recovers the entire spine in the leftmost column and delineates the fixation device in the middle column, which cannot be seen with the SE and VAT-SE sequences. The rightmost column shows the coronal reformats of the results; the SEMAC technique eliminates the several distortions, and reveals the four poles of the fixation device that are occluded by the metal-induced distortions with the SE and VAT-SE sequences. Note that small blurring exists in the coronal reformat of the SEMAC result, due to the prescription of the relative thick (4 mm) nominal slice thickness.

7
Comparison of in vivo results of the subject with a metallic fixation device in his spine. As compared to the SE (the first row) and the VAT-SE (the second row) sequences, the SEMAC technique (the bottom row) greatly suppresses the metal-induced distortions. ...

Figure 8 shows the in vivo results of the subject with several stainless steel screws in his lower leg. The results were obtained from the SE sequence (top row), the VAT-SE sequence (middle row) and the SEMAC technique (bottom row). With the SEMAC technique, the screws can be clearly visualized in both the in-plane results and the reformatted coronal view. In this study, the stainless steel screws result in such severe metal-induced field inhomogeneities that some resulting through-plane distortions fall outside the volume covered by the prescribed slices. The insufficient coverage of the spins leads to the signal loss, as evident by partial recovery of the rounded shapes of the drilled holes. In addition, the remaining uncorrected in-plane distortions manifest in the forms of ripple patterns (dotted arrows), which need to be addressed with stronger slice-select and/or readout gradients. Nevertheless, the SEMAC technique greatly suppresses the severe metal-induced distortions such that the images with diagnostic value are obtained.

8
Comparison of in vivo results of the subject with stainless steel screws in his lower leg. As compared to the SE (the first row) and the VAT-SE (the second row) sequences, the SEMAC technique (the bottom row) greatly suppresses the metal-induced distortions. ...

Discussion

The main difficulty in MRI near metallic implants results from large metal-induced field inhomogeneities, which vary with implant materials, implant sizes/shapes, and implant orientations with respect to main magnetic field [19]. As metal-induced field inhomogeneities cannot be accurately modeled in general, distortion-free MRI near metallic implants should rely on robust spatial encoding. A brute-force approach is to employ 3D phase encoding that incurs prohibitively long scan times [20]. In this work we propose an innovative imaging technique called SEMAC, which consists of a novel imaging sequence and a post-processing procedure to correct through-plane distortions.

The proposed SEMAC sequence, which provides robust encoding against metal-induced field inhomogeneities with a feasible scan time, is derived from several existing susceptibility imaging techniques. Specifically, spin-echo (SE) addresses the signal losses due to intra-voxel dephasing and the VAT-compensation gradient suppresses in-plane distortions. The additional z-phase encoding is closely related to z-shim technique used in functional MRI [21, 22] and gradient-echo slice excitation profile imaging (GESEPI) by Yang et al. [23]. In comparison to the z-shim technique that mainly deals with a linear field variation along the slice-select direction, the SEMAC technique corrects arbitrary field variations induced by metal. While GESEPI has been an effective means to recover the signal loss caused by intra-voxel dephasing in gradient-echo sequences, the SEMAC technique addresses large and arbitrary metal-induced field inhomogeneities that cause through-plane distortions across many slices.

Scan Parameters

The RF bandwidth in the SEMAC sequence is an important scan parameter that determines a tradeoff between suppressing in-plane distortions and resolving through-plane distortions. The VAT compensation gradient forces each excited spin to precess at the same Larmor frequency during the slice-selective excitation and the readout; hence, the metal-induced field inhomogeneities Δf are effectively compressed within the RF bandwidth BRF. As BRF [double less-than sign] Δf for imaging near metallic implants, the in-plane distortions are greatly suppressed. From Eq. 5, increasing the readout bandwidth per pixel Bp and/or reducing the RF bandwidth BRF can further improve the correction of in-plane distortions. However, a larger Bp leads to a loss in SNR, and a lower RF bandwidth leads to more severe through-plane distortions that require more slices to be acquired with a larger z-phase encoding FOV.

Through-plane distortions can only be fully corrected when all spins in a region-of-interest (ROI) are excited and spatially encoded. For an efficient coverage of the spins, we need to choose both nominal slice thickness s and the number of slices NS such that sNS is large enough to contain the largest through-plane distortion. The through-plane distortions are measured by reformatting the images obtained from the VAT-SE sequence in the (x, z) image plane. In general, for a given number of slices NS, the full coverage of all spins in the ROI should be provided while prescribing the thinnest possible nominal slice thickness. This is because thin nominal slice thickness requires a strong slice-select gradient, which reduces the through-plane distortions.

There are two scan parameters related to resolving excitation profiles, namely the number of z-phase encoding steps Nz and the z-phase encoding resolution δs. In the current implementation, the z-phase encoding resolution is set to the nominal slice thickness; i.e., δs = s. In this case, the number of z-phase encoding steps Nz should give a sufficiently large z-phase encoding FOV, which covers the through-plane distortions across ±Δfmax/BRF slices. Δfmax is the maximum metal-induced field inhomogeneity that can be measured with a spectroscopical pre-scan with all phase-encoding gradients and the readout gradient turned off. While setting δs = s is sufficient for most cases, we expect that the correction of through-plane distortions can be further improved with a higher z-phase encoding resolution. However, a higher z-phase encoding resolution requires more z-phase encoding steps to maintain the same z-phase encoding FOV, which leads to longer scan times.

Although the reported experiments are T1-weighted imaging performed on 1.5 T scanners, the SEMAC technique can be used to image near metallic implants at a high field strength (e.g., 3 T) and/or with T2 contrast. Since more severe distortions exist at a high field strength, a longer scan time is incurred to acquire more slices with a larger z-phase encoding FOV. As the SEMAC technique involves exciting multiple slices with no gap in between, the cross-talk between the slices reduces the effective TR, which in turn could affect T2 or proton density weighted imaging. The shortened effective TR can be addressed by designing RF pulses with exceptionally sharp slice profiles or prescribing longer repetition intervals. In addition, the SEMAC technique is compatible with extended echo train sequences, such as fast spin echo with extended echo train acquisition (FSE XETA) [24], which will lead to a more flexible tradeoff between contrast and scan times.

Acceleration Techniques

For the future work, to maintain feasible scan times or to achieve better correction of distortions, the SEMAC sequence will be incorporated with fast imaging techniques, such as partial k-space acquisition [25] along the phase-encoding y-axis. Exploiting the fact that the order of a multi-dimensional Fourier transform is interchangeable and the correction of through-plane and in-plane distortions are independent of the y-phase encoding, we can first apply an inverse Fourier transform along the slice-select direction to resolve the excitation profiles, followed by performing in-plane partial k-space reconstruction at each slice location. The accelerated data acquisition can also benefit from recent advances in parallel imaging [26, 27] and/or compressed sensing [28].

Another potential means to reduce scan times is to adapt the nominal thickness of each slice such that the largest through-plane distortion in each slice is kept roughly constant. Referring to Eq. 2, through-plane distortion is proportional to the product of field inhomogeneities and slice thickness. This suggests that a thicker slice can be prescribed at a location where field inhomogeneities are small. As long as the largest through-plane distortions in the slice do not exceed the corresponding z-phase encoding FOV, the excitation profile is correctly resolved without compromise in the resolution of z-phase encoding.

Conclusion

This paper presents an innovative MR imaging technique that effectively eliminates metal-induced distortions with feasible scan times. The key idea is to ensure that all spins in an ROI are excited once and are subject to robust spatial encoding against metal-induced field inhomogeneities. Both the phantom and in vivo results demonstrate the efficacy of the proposed SEMAC technique in obtaining distortion-free MR images near metallic implants. This technique holds great promise for significant clinical impact in evaluating millions of patients with metallic implants.

Acknowledgments

This work was supported by NIH-RR009784-11, NIH-EB002524-01, NIH-HL075803-01, NIH-CA077677-09, NIH 2P41RR009784-11, and Lucas Foundation.

Footnotes

A preliminary account of this work was presented at the 16th Meeting of the International Society for Magnetic Resonance in Medicine, May, 2008, page 838.

References

1. Kurtz S, Mowat F, Ong K, Chan N, Lau E, Halpern M. Prevalence of primary and revision total hip and knee arthroplasty in the united states from 1990 through 2002. J Bone Joint Surg Am. 2005;87:1487–1497. [PubMed]
2. Resnick D. Diagnosis of Bone and Joint Disorders. Saunders; New York: 2002.
3. Olsen RV, Munk PL, Lee MJ, Janzen DL, MacKay AL, Xiang QS, Masri B. Metal artifact reduction sequence: early clinical applications. Radiographics. 2000;20:699–712. [PubMed]
4. Eustace S, Goldberg R, Williamson D, Melhem ER, Oladipo O, Yucel EK, Jara H. MR imaging of soft tissues adjacent to orthopaedic hardware: techniques to minimize susceptibility artefact. Clin Radiol. 1997;52:589–594. [PubMed]
5. Ludeke KM, Roschmann P, Tischler R. Susceptibility artefacts in NMR imaging. Magn Reson Imaging. 1985;3:329–343. [PubMed]
6. Graf H, Steidle G, Martirosian P, Lauer UA, Schick F. Metal artifacts caused by gradient switching. Magn Reson Med. 2005;54:231–234. [PubMed]
7. Camacho CR, Plewes DB, Henkelman RM. Nonsusceptibility artifacts due to metallic objects in MR imaging. J Magn Reson Imaging. 1995;5:75–88. [PubMed]
8. Skare S, Anderson J. Correction of MR image distortions induced by metallic objects using a 3D cubic B-Spline basis set: application to stereotactic surgical planning. Magn Reson Med. 2005;54:169–181. [PubMed]
9. Butts K, Gold GE. Correction of slice profile distortions from metallic devices. Proceedings of the 14th Annual Meeting of ISMRM; Seattle. 2006. p. 2380.
10. Venook RD, Matter NI, Ramachandran M, Ungersma SE, Gold GE, Giori NJ, Macovski A, Scott GC, Conolly SM. Prepolarized magnetic resonance imaging around metal orthopedic implants. Magnetic Resonance in Medicine. 2006;56:177–186. [PubMed]
11. Cho ZH, Kim DJ, Kim YK. Total inhomogeneity correction including chemical shifts and susceptibility by view angle tilting. Medical Physics. 1988;15:7–11. [PubMed]
12. Butts K, Pauly JM, Gold GE. Reduction of blurring in view angle tilting mri. Magn Reson Med. 2005;53:418–24. [PubMed]
13. Kolind SH, MacKay AL, Munk PL, Xiang QS. Quantitative evaluation of metal artifact reduction techniques. J Magn Reson Imaging. 2004;20:487–495. [PubMed]
14. Koch KM, Busse RF, Lewein T, Potter HG, Hinks R, King K. Multiple resonant frequency offset acquisitions for imaging of metallic implants. Proceedings of the 16th Annual Meeting of ISMRM; Toronto. 2008. p. 1250.
15. Lu W, Butts K, Gold GE, Pauly JM, Hargreaves BA. Towards artifact-free MRI near metallic implants. Proceedings of the 16th Annual Meeting of ISMRM; Toronto. 2008. p. 838.
16. Koch KM, Hinks R, King K. Empirical and computed B0 perturbations induced by metallic implants. Proceedings of the 16th Annual Meeting of ISMRM; Toronto. 2008. p. 1180.
17. Nishimura DG. Principles of Magnetic Resonance Imaging. Stanford University; Stanford: 1996.
18. Speier P, Trautwein F. A calibration for radial imaging with large inplane shifts. Proceedings of the 13th Annual Meeting of ISMRM; Miami. 2005. p. 2295.
19. Guermazi A, Miaux Y, Zaim S, Peterfy C, White D, Genant HK. Metallic artefacts in MR imaging: effects of main field orientation and strength. Clin Radiol. 2003;58:322–328. [PubMed]
20. Graf H, Steidle G, Martirosian P, Lauer UA, Schick F. MRI of hip prostheses using single-point methods: in vitro studies towards the artifact-free imaging of individuals with metal implants. Magn Reson Imaging. 2004;22:1097–1103. [PubMed]
21. Glover GH. 3D z-shim method for reduction of susceptibility effects in BOLD fMRI. Magn Reson Med. 1999;42:290–299. [PubMed]
22. Constable RT, Spencer DD. Composite image formation in z-shimmed functional MR imaging. Magn Reson Med. 1999;42:110–117. [PubMed]
23. Yang QX, Williams GD, Demeure RJ, Mosher TJ, Smith MB. Removal of local field gradient artifacts in equation M19-weighted images at high fields by gradient-echo slice excitation profile imaging. Magnetic Resonance in Medicine. 1998;39:402–409. [PubMed]
24. Busse RF, Hariharan H, Vu A, Brittain JH. Fast spin echo sequences with very long echo trains: Design of variable refocusing flip angle schedules and generation of clinical t2 contrast. Magn Reson Med. 2006;55:1030–1037. [PubMed]
25. Noll DC, Nishimura DG, Macovski A. Homodyne detection in magnetic resonance imaging. IEEE Transactions on Medical Imaging. 1991;10:154–163. [PubMed]
26. Pruessmann KP, Weiger M, Scheidegger MB, Boesiger P. SENSE: Sensitivity encoding for fast MRI. Magnetic Resonance in Medicine. 1999;42:952–962. [PubMed]
27. Griswold MA, Jakob PM, Heidemann RM, Nittka M, Jellus V, Wang J, Kiefer B, Haase A. Generalized autocalibrating partially parallel acquisitions (GRAPPA) Magnetic Resonance in Medicine. 2002;47:1202–1210. [PubMed]
28. Lustig M, Donoho D, Pauly JM. Sparse MRI: The application of compressed sensing for rapid MR imaging. Magnetic Resonance in Medicine. 2007;58:1182–1195. [PubMed]