Alternating gradients for metal-induced artifacts correction in magnetic resonance imaging

Abstract

A method for magnetic resonance imaging suppresses off-resonance gradient-induced image artifacts due to metal. The method includes performing by a magnetic resonance imaging (MRI) apparatus two multi-spectral imaging (MSI) acquisitions within a field of view of the MRI apparatus, where the two MSI acquisitions have alternating-sign readout gradients. The two MSI acquisitions are then processed and combined by the MRI apparatus using a weighted image combination to produce a final image.

Claims

1. A method for magnetic resonance imaging that suppresses off-resonance gradient-induced image artifacts due to metal, the method comprising: performing by a magnetic resonance imaging (MRI) apparatus two multi-spectral imaging (MSI) acquisitions within a field of view of the MRI apparatus, where the two MSI acquisitions have alternating-sign readout gradients during signal readout; and processing and combining by the MRI apparatus the two MSI acquisitions using a weighted image combination to produce a final image; wherein combining the two MSI acquisitions using the weighted image combination comprises weighting a MSI acquisition m.sup.+ with weight w.sup.+, weighting a MSI acquisition m.sup. with weight w.sup., where the weights w.sup.+ and w.sup. include effects of both an RF excitation weight w.sub.RF and local gradient weight w.sub.G, and where MSI acquisition m.sup.+ and MSI acquisition m.sup. have opposite directions of readout gradients.

2. The method of claim 1 wherein the two MSI acquisitions have alternating-sign slice-select gradients and view-angle tilting gradients.

3. The method of claim 1 wherein the RF excitation weight w.sub.RF and the local gradient weight w.sub.G are determined from a composite field map f.

4. The method of claim 3 wherein the composite field map f is determined from MSI acquisition m.sup.+, MSI acquisition m.sup., and corresponding field maps f.sup.+ and f.sup..

Description

BRIEF DESCRIPTION OF THE SEVERAL VIEWS OF THE DRAWINGS

(1) FIG. 1A and FIG. 1B are MRI images with opposing readout directions, according to an embodiment of the invention.

(2) FIG. 1C is a combination of the two images of FIG. 1A and FIG. 1B, with pile-up artifacts suppressed, according to an embodiment of the invention.

(3) FIG. 1D is an image of an off-resonance field map, with regions of large off-resonance identified, according to an embodiment of the invention.

(4) FIG. 1E is an image the off-resonance gradient in the readout direction (G.sub.o), identifying regions of rapid off-resonance variations in the readout direction, according to an embodiment of the invention.

(5) FIG. 1F is a graph of the acquisition signal along the dashed vertical line in FIG. 1A, correlated with the effective readout gradient (G.sub.x+G.sub.o), according to an embodiment of the invention.

(6) FIG. 1G is a schematic diagram of an MRI apparatus, according to an embodiment of the invention.

(7) FIG. 2A is a sequence diagram used in an alternating-gradients acquisition, according to an embodiment of the invention.

(8) FIG. 2B is an illustration of a reconstruction processing pipeline that exploits the different locations of off-resonance-gradient-induced artifacts and non-excited regions between two alternating gradient acquisitions, according to an embodiment of the invention.

(9) FIG. 3A shows alternating-gradient acquisition images in opposite gradient directions and corresponding combined image for two slices, according to an embodiment of the invention.

(10) FIG. 3B shows computed local gradient and RF excitation weights in two opposite gradient directions, according to an embodiment of the invention.

(11) FIG. 4 shows results of the alternating-gradient technique in a volunteer with a total knee replacement, according to an embodiment of the invention.

(12) FIG. 5 shows results of the alternating-gradient technique in a patient with a total hip replacement, according to an embodiment of the invention.

(13) FIG. 6 shows results of the alternating-gradient technique with additional acceleration in a patient with a total hip replacement, according to an embodiment of the invention.

DETAILED DESCRIPTION OF THE INVENTION

(14) In state of the art MRI imaging, severe off-resonance-gradient-induced artifacts, including pile-ups and ripples, appear where the magnetic field gradient due to field inhomogeneities or off-resonance gradient G.sub.o is opposite in sign to the readout gradient G.sub.x. As a result, the effective readout gradient G.sub.o+G.sub.x in these areas has reduced magnitude. This local reduction in the magnitudes of the readout gradient expands the encoded pixel size, causing irrecoverable loss of resolution in the readout direction. This can also be viewed as a decreased k.sub.x traversal extent. Conversely, where the magnetic field gradient due to field inhomogeneities or off-resonance gradient G.sub.o has the same sign as the readout gradient G.sub.x, the effective readout gradient G.sub.o+G.sub.x magnitude increases. This additive effect reduces the encoded pixel size, but this effect can be mostly corrected by deblurring and Jacobian-based intensity correction. In two acquisitions, one with the readout gradient G.sub.x inverted with respect to the other, these artifacts appear in different locations. By appropriate combination of two such acquisitions pile-up/ripple artifacts can be suppressed.

(15) FIGS. 1A-F demonstrate the correspondence of off-resonance-gradient-induced artifacts and effective readout gradient magnitudes with a simulated MAVRIC-SL acquisition in a digital metal-implant phantom. Specifically, FIG. 1A and FIG. 1B are MRI images with opposing readout directions, and FIG. 1C is the combination of these two images using the techniques of the present invention, with pile-up artifacts suppressed. The localized regions 100, 102, 104, 106, 108 are off-resonance-gradient-induced artifacts including pile-ups and ripples, and blurring of the resolution grid. FIG. 1D is an image of the corresponding off-resonance field map, with localized off-resonance regions 110, 112, 114, 116. FIG. 1E is an image the off-resonance gradient in the readout direction (G.sub.o), showing off-resonance regions 118, 120, 122, 124, 126, 128. The graph of FIG. 1F shows the corresponding signal along the dashed vertical line in FIG. 1A, correlated with the effective readout gradient (G.sub.x+G.sub.o). The artifacts in these images appear in areas where local off-resonance gradient cancels the applied readout gradient (e.g., region 102 and 122, corresponding to the peak of signal 130).

(16) FIG. 1G is a perspective view of a conventional MRI apparatus, showing enclosure 132, field of view 136, and gantry 134. According to embodiments of the present invention, the MRI apparatus is operated as follows. The field of view is excited using alternating-gradients acquisition, e.g., using a sequence diagram such as the one shown in FIG. 2A. The diagram shows the signals for RF 200, and signals 202, 204, 206 for gradients G.sub.z, G.sub.y, G.sub.x, in three directions. The solid lines indicate the signals used for acquisition in one gradient direction, while the dashed lines indicate the signals used for acquisition in the opposite gradient direction. The voxel shearing effect caused by view-angle tilting (VAT), used with slice-selective MSI, can be matched between the two gradient directions by additionally inverting the slice-select and VAT gradients G.sub.z, which also changes the regions of non-excited signal loss.

(17) A reconstruction processing pipeline that exploits the different locations of off-resonance-gradient-induced artifacts and non-excited regions between the two acquisitions is illustrated in FIG. 2B. The locations of off-resonance-gradient-induced artifacts are predicted based on the estimated effective readout gradient magnitude |G.sub.o+G.sub.x|, and the non-excited regions are predicted based the composite frequency profile and estimated off-resonance frequency. In block 208, acquisition is performed in a first gradient direction. Similarly, in block 210, acquisition is performed in a second gradient direction, opposite to the first direction. In blocks 212 and 214, standard field map estimation, deblurring, and Jacobian-based intensity correction are performed separately for each of the acquired images from the first and second gradient directions. The resulting field maps and composite images are denoted as f.sup.+, f.sup., and m.sup.+, m.sup., where superscripts + and denote the different gradient directions.

(18) In block 216, a composite field map f is obtained by combining the two field maps, as follows:
f=[(m.sup.+).sup.2f.sup.++(m.sup.).sup.2f.sup.]/[m.sup.+).sup.2+(m.sup.).sup.2][Eq. 1]

(19) In block 220, the off-resonance gradient G.sub.o is computed as the finite difference off along the readout direction. The local gradient weights of each gradient direction are computed as

(20) $\begin{array}{cc}{w}_G^{}\left(x,y,z\right)=\max \left\{\min \left\{1\frac{G_o\left(x,y,z\right)}{G_x},1\right\},0\right\}& \left[\mathrm{Eq}.2\right]\end{array}$
where lower values indicate smaller magnitude of effective readout gradient and thus more severe off-resonance-gradient-induced artifacts.

(21) In block 218, RF excitation weights are computed as

(22) $\begin{array}{cc}{w}_{\mathrm{RF}}^{}\left(x,y,z\right)=\underset{b}{.Math.}{R}_b^2\left(f\left(x,y,z\right)\frac{}{2}{G}_z.Math.z\right)& \left[\mathrm{Eq}.3\right]\end{array}$
where the sum is over all bins b, R.sub.b() represents RF frequency profile of bin b, and represents the gyromagnetic ratio. Lower values indicate non-excited regions. In some embodiments, the method uses slice-selective MSI (e.g. MAVRIC-SL, SEMAC). Note that the alternating-gradient techniques of the present invention can also be applied to non-slice-selective MSI. For non-slice-selective MSI, G.sub.z=0. Consequently, only the readout gradient G.sub.x needs to be inverted in this case. In the combination scheme, the step of computing the RF excitation weights can be skipped since the excited regions are the same between two gradient directions. Thus, for embodiments using non-slice-selective MSI (e.g. MAVRIC), the slice-select/VAT gradient in FIG. 2A and the step 218 of computing excitation profiles and w.sub.RF can be skipped.

(23) In block 222, the weighted image combination is computed as

(24) $\begin{array}{cc}m=\frac{w^{+}.Math.m^{+}+w^{-}.Math.m^{-}}{w^{+}+w^{-}}& \left[\mathrm{Eq}.4\right]\end{array}$
where the overall weights
w.sup.+=exp{w.sub.RF.sup.++w.sub.G.sup.+}, w.sup.=exp{w.sub.RF.sup.+w.sub.G.sup.}[Eq. 5]
include the effects of both RF excitation weights and local gradient weights. The scaling factors and can be selected empirically.

(25) Images illustrating the alternating-gradient technique for a hip implant phantom is shown in FIGS. 3A-B.

(26) FIG. 3A is an image grid where columns 308, 310 correspond to acquired images in opposite first and second gradient field directions, and column 312 corresponds to the combined image using the alternating-gradient techniques of the present invention. Rows 314 and 316 correspond to two different acquisition slices. The artifact regions 300, 302, 304, 306 are off-resonance-gradient-induced signal variations and blurring, which are suppressed in the combined image.

(27) FIG. 3B is an image grid where the two rows show, respectively, computed local gradient weights and computed RF excitation weights for slice 316. The two columns correspond to the two gradient field directions 308, 310. The localized regions 318, 320, 322, 324, 326, 328 of the gradient weights match well with the regions of off-resonance-gradient-induced artifacts. The regions 330, 332, 334, 336 of the RF excitation weights indicate non-excited regions of each direction. The data is from a MAVRIC-SL acquisition on a phantom with a total hip replacement and a resolution grid. The following scan parameters were used: B.sub.0=3T, matrix size=38419224, voxel size=0.81.33.0 mm.sup.3, scan time=7.7 min.

(28) The alternating-gradients acquisition and combination techniques of the present invention can be applied with both slice-selective and non-slice-selective MSI sequences for suppressing off-resonance-gradient-induced artifacts in imaging of various metallic implants. The method can reduce the artificial intensity variations and recover the lost resolution to improve the image quality in close vicinity of metal. Two examples demonstrating the method with MAVRIC-SL are shown in FIG. 4 and FIG. 5.

(29) FIG. 4 shows results of the alternating-gradient technique in a volunteer with a total knee replacement with readout direction superior/inferior. The three columns correspond to three coronal slices on the anterior side, near the center, and on the posterior side, respectively. The dotted arrows 400, 408, 412 point to areas where more signal loss was observed in one gradient direction. The solid arrows 402, 404, 406, 410 point to off-resonance-gradient induced signal variations in images of individual gradient directions.

(30) The following scan parameters were used: B.sub.0=3T, matrix size=25625624, voxel size=0.70.64.0 mm.sup.3, scan time=6.9 min.

(31) FIG. 5 shows results of the alternating-gradient technique in a patient with a total hip replacement. The two rows correspond to different slices. Columns 500 and 502 are acquired images in opposite gradient field directions, detailing regions 506 and 508, respectively. The regions 514, 516, 518, 520 are off-resonance-gradient-induced signal variations in images of individual gradient directions, which were suppressed in the combined images, shown in column 504. It is clear that the combined image recovers substantial signal around the head of the implant that is lost in the individual images. Reformatted sagittal images in different gradient directions across the implant head are shown on the left 512, where the regions 522, 524, 526, 528 point to signal loss. The following scan parameters were used: B.sub.0=3T, matrix size=38425640, voxel size=1.01.54.0 mm.sup.3, scan time=6.0 min.

(32) In light of the teachings of the present invention, those skilled in the art will appreciate that the principles of the invention are not limited to the specific examples described above for purposes of illustration. Several variations of the technique are possible and envisioned by the inventors.

(33) For example, methods to accelerate the alternating-gradient acquisition can be integrated to the techniques of the present invention to shorten the scan time, including acceleration methods for general MRI acquisitions and specific for MSI.

(34) The total scan time of the two alternating-gradient acquisitions was equal to the conventional scan. A model-based reconstruction method (9) was used to reconstruct the bin images from prospectively under-sampled data.

(35) FIG. 6 compares conventional imaging results with results of the alternating-gradient technique with additional 2 acceleration in a patient with a total hip replacement. The two rows correspond to two different slices. Column 600 shows final combined images using the alternating-gradient technique, while column 602 shows the images for a conventional MRI scan. The images in these columns are detail of the regions 604 and 606 in the leftmost column. The alternating-gradient acquisitions were prospectively under-sampled by a factor of 2 in addition to ARC and half-Fourier, and were reconstructed by the model-based reconstruction method. The conventional acquisition was the product MAVRIC-SL sequence (single gradient direction, ARC and half-Fourier sampling). Therefore, their total scan times were the same. The regions 608, 610, 612, 614, 616, 618, 620, 622, 624 are regions where off-resonance-gradient-induced signal variations were present in the acquired images. It is clear that the artifacts were suppressed using the alternating gradient technique (column 600) but not in the conventional technique (column 602). The structures in these areas, which were originally obscured by the artifacts, were revealed in the images of column 600. The following scan parameters were used: B.sub.0=3T, matrix size=38425648, voxel size=1.01.54.0 mm.sup.3, scan time=7.2 min.

(36) Methods to exploit redundancies in images of the two gradient directions can be added to the proposed method to allow under-sampling of data and thus shorten the scan time. For example, in most part of the FOV away from the metallic implants, the images of the two gradient directions could be the same. This constraint could be enforced in the image reconstruction algorithm to suppress aliasing artifacts due to under-sampling of k-space data.

(37) In MSI, multiple acquisitions are performed with different center frequencies to image spins across a wide range of off-resonance frequencies, and each acquisition is usually referred to as a spectral bin. When performing alternating-gradient acquisitions, spectral bins of two gradient directions can be acquired in an interleaved fashion to reduce the influence of inter-scan motion. Specifically, the acquisition order can be: spectral bin 1 of direction 1, spectral bin 1 of direction 2, spectral bin 2 of direction 1, spectral bin 2 of direction 2, etc.

(38) The computation of the overall weights combining the local gradient weights and RF excitation weights can be modified to suppress artifacts better and to make transitions smoother in the combined image. In the technical description, the multiplications of exponential of local gradient and RF excitation weights are used as the overall weights to linearly combine the images of both directions. Other combination methods can also be used in substitution of the above method. An alternative combination scheme is as follows: step 1, for each voxel, the gradient direction of higher overall weight (computed following Eq. 5) is given weight 1, the other gradient direction is given weight 0, resulting in one binary weight map for each direction; step 2, the weighting maps from step 1 are smoothed around the edges to avoid discontinuities in the combined image. This combination scheme avoids averaging the images of different gradient directions in most part of the FOV and possible blurring due to the averaging operation.

(39) Other metrics of image quality of individual gradient directions can be integrated to the overall weights. For example, gradient entropy can be used to evaluate the level of motion artifacts in images of individual directions, and suppress motion artifacts in the combined image.