Compensation of magnetic field inhomogeneity in MR spectroscopic imaging using dynamic k-space expansion in combination with parallel imaging

Abstract

A method for the compensation of magnetic field inhomogeneity in magnetic resonance spectroscopic imaging comprising the steps of using dynamic k-space expansion in combination with parallel imaging.

Claims

1. A method for the compensation of magnetic field inhomogeneity in magnetic resonance spectroscopic imaging comprising the steps of using dynamic k-space expansion in combination with parallel imaging; said k-space is expanded with increasing spectral encoding time t, resulting in spectral line narrowing in proportion to the expansion of k-space; and minimizing gradient switching by tailoring the expansion and density of k-t-space sampling to the dispersion and density of signal trajectories in k-t-space.

2. The system and method of claim 1 wherein said expansion of the k-space with spectral encoding time includes interleaving progressively larger spatial encoding gradient moments.

3. The method of claim 1 wherein expanding the k-space is accomplished by extending high speed image encoding modules, including echo-planar and spiral encoding modules.

4. The method of claim 1 wherein said k-space is undersampled regularly and compressed sensing is used to reconstruct the missing data.

5. The method of claim 1 wherein said k-space is under sampled randomly and compressed sensing is used to reconstruct the missing data.

6. The method of claim 1 wherein said expansion of k-t-space is linear and employs readout gradient moment with stepwise increases (2G.sub.lδt) every second gradient using a constant gradient duration δ, up to the limits of the gradient performance.

7. The method of claim 1 wherein single-shot phase encoding using gradient blips with linearly increasing gradient moment G.sub.l*t are selectively interleaved into the readout, with a corresponding increase of the effective spectral dwell time.

8. The method of claim 7 wherein said interleaving starts at the edges of the original k.sub.y-k.sub.z-space and progressively inserts single-shot phase encoding into more central k.sub.y-k.sub.z-space encodings as time t increases.

9. The method of claim 1 wherein the k-space dimensions are tailored to the orientation and amplitude distribution of local Gradients G.sub.l based on B.sub.0 gradient maps.

10. The method of claim 1 further including the step of compensating local gradients in a selected brain region and simultaneously acquiring signals from the rest of the brain without compensation.

11. A method for the compensation of magnetic field inhomogeneity in magnetic resonance spectroscopic imaging comprising the steps of using dynamic k-space expansion in combination with parallel imaging; and said k-space is expanded with increasing spectral encoding time t, resulting in spectral line narrowing in proportion to the expansion of k-space; and wherein said k-space is undersampled regularly and partial parallel imaging is used to reconstruct the missing data.

12. A method for the compensation of magnetic field inhomogeneity in magnetic resonance spectroscopic imaging comprising the steps of using dynamic k-space expansion in combination with parallel imaging; and said k-space is expanded with increasing spectral encoding time t, resulting in spectral line narrowing in proportion to the expansion of k-space; and wherein said k-space is under sampled randomly and compressed sensing is used to reconstruct the missing data.

Description

BRIEF DESCRIPTION OF THE SEVERAL VIEWS OF THE DRAWINGS

(1) In the drawings, which are not necessarily drawn to scale, like numerals may describe substantially similar components throughout the several views. Like numerals having different letter suffixes may represent different instances of substantially similar components. The drawings illustrate generally, by way of example, but not by way of limitation, a detailed description of certain embodiments discussed in the present document.

(2) FIG. 1 shows an example of local group spin echo shifts in k-t-space where local gradients progressively move the signal maximum (phase encoding echo) outside of data space and a shorter data trace in the time domain corresponds to a filter and causes line broadening.

(3) FIG. 2A shows a single-shot spectroscopic imaging method that compensates local gradients in regions that suffer from susceptibility related spectral line broadening.

(4) FIG. 2B depicts the two distinct linear k-space trajectories that are encoded using the interleaved shimming method described in FIG. 2A.

(5) FIG. 3 shows an example of linear expansion of k-t-space encoding to capture the expansion of local group spin echo shifts along the k-t-space axes.

(6) FIG. 4A provides an example of linear expansion of k-t-space encoding in 3 time slices for sections delineating k-t-space regions that are encoded using interleaved phase encoding.

(7) FIG. 4B provides an example of linear expansion of k-t-space encoding in 3 time slices for a 2D PEPSI readout module with 2×2-fold k-space expansion using a linear increase in readout gradient moment and interleaved phase encoding gradients blips as a function of conventional phase encoding step (k.sub.y) and t.

(8) FIG. 4C shows zero filling of k-space depicted by empty sections of bounding boxes to obtain consistent data sizes across time slices.

(9) FIGS. 5A and 5B show a PEPSI pulse sequence simulation of excitation and readout modules with the linear expansion of k-t-space encoding using 6 different readout gradient train segments with increasing readout gradient and phase encoding gradient moments, and corresponding increases in ADC duration at constant dwell time.

(10) FIGS. 6A, 6B and 6C show a corresponding PEPSI pulse sequence simulation with the nonlinear expansion of k-t-space encoding using 6 different readout gradient train segments with nonlinearly increasing readout gradient and phase encoding gradient moments, and corresponding increases in ADC duration at constant dwell time.

(11) FIGS. 7A and 7B show a PEPSI pulse sequence simulation of excitation and readout modules with linear expansion of k-t-space encoding using 6 different readout gradient train segments. The simulation shows the first phase encoding step with maximum positive k.sub.y encoding moment (see Y Gradient and Y-Gradient moment).

(12) FIGS. 8A and 8B show a corresponding simulation of the central phase encoding step with minimum k.sub.y encoding moment (see Y Gradient and Y-Gradient moment).

(13) FIGS. 9A and 9B show a corresponding simulation of the central phase encoding step with maximum negative k.sub.y encoding moment (see Y Gradient and Y-Gradient moment).

(14) FIGS. 10A, 10B and 10C show a corresponding simulation of all phase encoding steps, including water suppression modules that precede the excitation and readout modules.

(15) FIG. 11A shows a 2D K-space expansion of time slices in different readout segments using a combination of full k-space encoding and undersampled encoding for a full k-space encoding in segment 1.

(16) FIG. 11B shows a 2D K-space expansion of time slices in different readout segments using a combination of full k-space encoding and undersampled encoding with for a 4-fold expansion of full k-space encoding in segment 2 using 2-fold increased readout gradient moment and blipped phase encoding gradients interleaved between positive and negative readout gradients to encode outer k.sub.y-space lines.

(17) FIG. 11C shows a 2D K-space expansion of time slices in different readout segments using a combination of full k-space encoding and undersampled encoding for a 9-fold expansion of k-space encoding in segment 4 using 3-fold increased readout gradient moment and blipped phase encoding gradients interleaved between positive and negative readout gradients to encode outer k.sub.y-space lines with 2-fold undersampling.

(18) FIG. 11D shows a 2D K-space expansion of time slices in different readout segments using a combination of full k-space encoding and undersampled encoding for a 16-fold expansion of k-space encoding in segment 4 using 4-fold increased readout gradient moment and blipped phase encoding gradients interleaved between positive and negative readout gradients to encode outer k.sub.y-space lines with 3-fold undersampling. Subsequent segments are acquired with correspondingly larger outer k.sub.y undersampling factors.

(19) FIG. 12A shows a 2D K-space expansion of time slices in different readout segments using full k-space encoding for a full k-space encoding in segment 1.

(20) FIG. 12B shows a 2D K-space expansion of time slices in different readout segments using full k-space encoding for a 4-fold expansion of full k-space encoding in segment 2 using 2-fold increased readout gradient moment and blipped phase encoding gradients interleaved between positive and negative readout gradients to encode outer k.sub.y-space lines.

(21) FIG. 12C shows a 2D K-space expansion of time slices in different readout segments using full k-space encoding for a 9-fold expansion of k-space encoding in segment 4 using 3-fold increased readout gradient moment and 3 blipped phase encoding gradients interleaved between positive and negative readout gradients to encode outer k.sub.y-space lines.

(22) FIG. 12D shows a 2D K-space expansion of time slices in different readout segments using full k-space encoding for a 16-fold expansion of k-space encoding in segment 4 using 4-fold increased readout gradient moment and 4 blipped phase encoding gradients interleaved between positive and negative readout gradients to encode outer-space lines. Subsequent segments are acquired with a correspondingly larger number of interleaved gradient blips.

(23) FIG. 13 shows a 3D K-space expansion of time slices in different readout segments depicted in the k.sub.y-k.sub.z plane.

(24) FIGS. 14A, 14B and 14C show a simulation of spectral reconstruction of undersampled data.

(25) FIG. 15A shows a GRAPPA k-space reconstruction of k.sub.y-undersampled raw data in a spherical phantom using OPENGRAPPA for an undersampled k-space data at selected time points at the beginning of the 6 acquired readout segments for a selected coil located in the vicinity of a susceptibility inhomogeneity due to an air bubble.

(26) FIG. 15B shows a GRAPPA k-space reconstruction of k.sub.y-undersampled raw data in a spherical phantom using OPENGRAPPA for GRAPPA reconstructed k-space data in the k.sub.y-undersampled readout segments 3-6 replacing the missing data lines in the outer segments of ky space.

(27) FIG. 16A shows a phantom experiment for an MRI with ROIs indicating magnetically homogeneous region and magnetically inhomogeneous region.

(28) FIG. 16B shows a phantom experiment for an undersampled spectrum with k-space expansion from the central magnetically uniform region reconstructed using the expanded Fourier transform.

(29) FIG. 16C shows a phantom experiment fora corresponding spectrum reconstructed online from fully sampled data without k-space expansion using conventional PEPSI.

(30) FIG. 16D shows a phantom experiment for a spectrum without k-space expansion from the magnetically inhomogeneous region close to the air bubble shows a broad water peak.

(31) FIG. 16E shows a phantom experiment for a corresponding water spectrum with k-space expansion shows a narrower spectral line width.

(32) FIG. 17A shows a 2D PEPSI data acquired in a healthy control with 2D K-space expansion for an anatomical localizer with sagittal PEPSI slice prescription and outer volume suppression slices.

(33) FIG. 17B shows a 2D PEPSI data acquired in a healthy control with 2D K-space expansion for a first time slice of water suppressed data maps, which maps residual water with a strong signal in inferior frontal cortex that corresponds to a brain region with strong magnetic field inhomogeneity and local frequency shifts in which the water suppression efficiency was reduced.

(34) FIG. 17C shows a 2D PEPSI data acquired in a healthy control with 2D K-space expansion for a threshold ratio map of the residual signal in the first w.r.t. the last time slice for the acquisition without k-space expansion that shows a particularly strong signal loss in inferior brain regions due to magnetic field inhomogeneity (threshold: 40).

(35) FIG. 17D shows a 2D PEPSI data acquired in a healthy control with 2D K-space expansion for a corresponding ratio for acquisition with 6×6-fold k-space expansion shows considerably reduced signal loss in most brain regions except centrum semiovale.

(36) FIG. 17E shows a 2D PEPSI data acquired in a healthy control with 2D K-space expansion for K-space raw data in the first time slice of the acquisition with 6×6-fold k-space expansion for coil 24 in, which is adjacent to the frontal cortex.

(37) FIG. 17F shows a 2D PEPSI data acquired in a healthy control with 2D K-space expansion for corresponding k-space raw data in the last time slice, which illustrates the considerable dispersion of signals along k.sub.x and k.sub.y.

DETAILED DESCRIPTION OF THE INVENTION

(38) Detailed embodiments of the present invention are disclosed herein; however, it is to be understood that the disclosed embodiments are merely exemplary of the invention, which may be embodied in various forms. Therefore, specific structural and functional details disclosed herein are not to be interpreted as limiting, but merely as a representative basis for teaching one skilled in the art to variously employ the present invention in virtually any appropriately detailed method, structure, or system. Further, the terms and phrases used herein are not intended to be limiting, but rather to provide an understandable description of the invention.

(39) Compensation of Local Gradients by Dynamic Case k-Space Expansion

(40) In one embodiment, the present invention provides for the global compensation of spectral line broadening in multiple regions with differing local gradient vectors. This may be accomplished by expanding the k-space with increasing spectral encoding time t, resulting in spectral line narrowing in proportion to the expansion of k-space. This k-space expansion approach does not increase spatial encoding time and it is more SNR efficient than approaches that increase spatial resolution.

(41) Increasing spatial resolution requires uniform expansion of k-space for all time slices, thus penalizing early time slices. However, expansion of k-space with spectral encoding time requires interleaving progressively stronger spatial encoding into the spectral acquisition and consequent elongation of the spectral dwell time, which decreases spectral bandwidth.

(42) The resulting undersampling, which increases with spectral encoding time is feasible, however, since spectral information density in k-t-space is typically sparse and decreases with increasing spectral encoding time. Spectral reconstruction of nonuniformly sampled data is performed by applying either non-uniform fast Fourier transform (NUFFT) regridding or expanded Fourier Transform, a MATLAB toolbox (https://www.mathworks.com/matlabcentral/fileexchange/11020-extended-dft). For example, the proton spectrum is quite sparse at long time delays and contains only a few peaks (water, Ino, Cho, Cr, NAA and perhaps lipids and lactate), which facilitates this approach. Spectral reconstruction of nonlinearly sampled data is well established using the non-linear Fourier Transform.

(43) This k-space expansion approach is compatible with conventional phase encoded MRSI and particularly suitable for high-speed MRSI, such as echo-planar MRSI and spiral MRSI. With echo-planar encoding, increasing the readout gradient moment with spectral encoding time expands kz and interleaving blipped phase encoding gradients (e.g. using single-shot interleaved phase encoding) expands k.sub.y and k.sub.z as shown in FIG. 3. Specifically, FIG. 3 is a generalization, which compensates local gradients with arbitrary orientations and a range of amplitudes with a maximum amplitude that corresponds to the rate of change in k-space extent.

(44) With spiral MRSI, expanding k-space can be accomplished by extending the spiral encoding module. More complex multi-axis gradient waveforms (e.g. to encode spherical trajectories) or combinations thereof, and switched nonlinear surface gradients, may also be particularly effective for dynamic k-space expansion. As gradient moments increase to encode the expanding k-space it becomes necessary to increase the spectral dwell time to accommodate the increasingly longer gradient encoding modules, leading to a decrease of the sampled spectral width, which must still fulfill the Nyquist criterion for sampling a minimum spectral width that fully resolves aliased spectral peaks.

(45) A preferred implementation that minimizes gradient switching tailors the expansion and density of k-t-space sampling to the dispersion and density of signal trajectories in k-t-space, which can be directly predicted from B.sub.0 mapping.

(46) Combining Gradient Encoding with Partial Parallel Imaging and Compressed Sensing

(47) The expansion of k-space may necessitate progressive k-space undersampling to maintain a desired maximum spectral dwell time. To relax the requirements of gradient encoding, in another embodiment, the present invention undersamples k-space either regularly or randomly, and uses partial parallel imaging or compressed sensing to reconstruct the missing data. The combination of high-speed MRSI with partial parallel imaging using GRAPPA and SENSE, and compressed sensing, has been introduced. In other aspects, the present invention samples the initial time slices without undersampling to enable computation of the reconstruction k-space kernel in case of GRAPPA or the sensitivity profiles for image space unfolding in case of SENSE. In still further aspects, the present invention is configured to increasingly undersample k-space with increasing spectral encoding time to maintain a desired maximum spectral dwell time.

(48) Dynamic K-Space Expansion in PEPSI Using Increasing Readout Gradient and Interleaved Phase Encoding Gradient Moments

(49) In other aspects of the present invention, the linear expansion of k-t-space may use a readout gradient moment with stepwise increases (2G.sub.lδt) every second gradient using a constant gradient duration δ, up to the limits of the gradient performance. Single-shot phase encoding using gradient blips with linearly increasing gradient moment G.sub.l*t are selectively interleaved into the PEPSI readout, with a corresponding increase of the effective spectral dwell time. To minimize the SNR loss in magnetically homogeneous areas, the interleaving will start at the edges of the original k.sub.y-k.sub.z-space and progressively insert single-shot phase encoding into more central k.sub.y-k.sub.z-space encodings as time t increases, as shown in FIGS. 4A-4C. In FIG. 4A, sections 400 and 410 delineate k-t-space regions that are encoded using interleaved phase encoding.

(50) The k-space dependent time delay Δt of this insertion is Δt=T (k.sub.max−k)/k.sub.max, where Tis the total readout duration and k.sub.max is the extent of the original k-space. To maximize SNR, the extent of k-t-space expansion along the different k-space dimensions is tailored to the orientation and amplitude distribution of local Gradients G.sub.l based on B.sub.0 gradient maps. The SNR scales to at least with the square root of the decrease in voxel size in regions with magnetic field inhomogeneity depending on the histogram of local magnetic field gradients in the volume of interest. This k-t-space expansion method provides a reduction in line width that is comparable to that of increasing spatial resolution, however, with much improved SNR and without increase in scan time. A √{square root over (2)} larger gain in SNR may be obtained in magnetically homogeneous brain regions.

(51) Reconstruction

(52) Zero-filling of time-slice data in the k-space domain is performed as a first step to obtain consistent k-space matrix size across time slices as shown in FIG. 4C. Spectral reconstruction is performed by applying non-uniform fast Fourier transform (NUFFT) regridding or the expanded Fourier Transform MATLAB toolbox. Conventional Fourier-based spatial reconstruction is performed either before or after spectral reconstruction. This k-t-space expansion method is compatible with partial parallel imaging using SENSE and GRAPPA.

(53) Practical Implementation of a PEPSI Pulse Sequence with Dynamic k-Space Expansion

(54) A pulse sequence was implemented on a Siemens Trio scanner (Syngo VB17A) using step-wise increases in readout gradient moments that were realized using readout gradient train segments with increasing readout gradient duration at constant ADC readout bandwidth per pixel. Interleaved alternating gradients are switched along the k.sub.y and k.sub.z axes to encode multiple k.sub.x lines in a single shot. The moments of these phase encoding gradient blips increase from segment to segment to expand k.sub.y-space. The pulse sequence implementation provides flexible control of readout gradient moments, duration of readout gradient train segments, and phase encoding gradient blip moments for each readout gradient train segment and for each phase encoding step, using a combination of mathematical expressions coded in C++ inside the pulse sequence, GUI based parameter selection and a lookup table in form of an external text file to maximize flexibility. An example of the excitation and readout modules of a PEPSI pulse sequence with linear k-space expansion and 2 k.sub.x lines acquired per phase encoding step is shown in FIGS. 5A-5B. An example of the corresponding PEPSI pulse sequence with nonlinear k-space expansion is shown in FIGS. 6A-6C. FIGS. 7-10 depict simulations of different phase encoding steps of the entire PEPSI pulse sequence with embedded water reference and navigators for the case of linear k-space expansion. The corresponding k-space encoding of time slices in different readout segments is shown in FIGS. 11A-11D where full k-space encoding is shown in sections 1100A-1100D and undersampled encoding is shown in sections 1110A-1110D. Note that k.sub.x is fully sampled. Central sections of k.sub.y-space are fully sampled as well, whereas peripheral sections of k.sub.y are undersampled.

(55) FIGS. 12a-12d depict an example of expanding k-space encoding with full sampling of both k.sub.x and k.sub.y, which requires interleaving of additional phase encoding blips to acquire more than 2 k-space lines in a single readout module at the expense of increasing spectral dwell time. Full k-space encoding is used along k.sub.y-k.sub.z plane kx and in the center of the k.sub.y-k.sub.z planes 1200A-1200D and undersampled k-space encoding 1210A and 1210B in the outer regions of the k.sub.y-k.sub.z plane increases linearly from readout segment to readout segment (numbered: 1,2,3 . . . ) using increasing undersampling factors.

(56) FIG. 13 depicts an example of expanding k-space encoding with full sampling of k.sub.x and increasing undersampling in the k.sub.x-k.sub.y-plane in consecutive readout segments. Full k-space encoding is used along k.sub.y-k.sub.z plane kx and in the center of the k.sub.y-k.sub.z plane 1300A-1300D and undersampled k-space encoding 1310A-1310C in the outer regions of the k.sub.y-k.sub.z plane increases linearly from readout segment to readout segment (numbered: 1,2,3 . . . ) using increasing undersampling factors.

(57) FIGS. 14A-14C demonstrate that temporally undersampled data can be reconstructed with a minor increase in spectral line width using the expanded Fourier Transform MAT LAB toolbox (https://www.mathworks.com/matlabcentral/fileexchange/11020-extended-dft). FIG. 14A shows the simulated time domain data set with temporal undersampling that increases from 2-fold in the 2.sup.nd segment to 6-fold in the 6th segment. FIG. 14B shows the reconstructed spectrum using the expanded Fourier Transform. Undersampling increases the spectral line width linewidth slightly with respect to the fully sampled data as shown in FIG. 14C but does not distort the spectral pattern. A preferred implementation of k-space expansion employs much finer discretization steps.

(58) Data were acquired in a spherical phantom containing metabolites and in a healthy control using a Siemens Trio 3 Tesla MRI scanner equipped with 32-channel head array RF coil. Water reference and water suppressed data were acquired in a sagittal or axial slice using TR/TE=2200/15 ms, 32×32 spatial matrix, 8×8×20 mm.sup.3 nominal voxel size and 1:24 min scan time. Water reference data were acquired with 150 readout gradients. Water-suppressed data were acquired with 256 readout gradients. The readout was along the z-axis. Six equidistant readout segments with linearly increasing k-space expansion (1×, 2×, 3×, 4×, 5× and 6×, along both k.sub.x and k.sub.y) were used. Undersampling of k.sub.y-encoding increased from 2-fold in the 3.sup.rd segment to 5-fold in the 6th segment. Accordingly, the acquired spatial data matrix in the 6th readout segment was 392×64 complex data points (readout×phase encoding) and time slice undersampling increased from 2-fold in the 2.sup.nd segment to 6-fold in the 6.sup.th segment. Data were reconstructed offline using zero filling of the readout direction to obtain a consistent matrix size of 392×64 data points (readout×phase encoding) for all time slices. FIGS. 15A-15B show k-space raw data acquired in the phantom using increasing k.sub.y-undersampling in readout segments 3-6 and k-space reconstructed data using the OPENGRAPPA. FIGS. 16A-16E display localized spectra in magnetically uniform and inhomogeneous regions of the phantom, reconstructed from undersampled data using the expanded Fourier Transform, that illustrate the improved spectral linewidth obtained with k-space expansion. FIG. 16A shows magnetically homogeneous region 1600 and magnetically inhomogeneous region 1610. FIGS. 17A-17F summarize the in vivo data and shows that the signal decrease in most brain regions is strongly reduced with k-space expansion. The signal gain at the last time slice, which corresponds to an acquisition delay of 500 ms, ranges from 3-fold in occipital and parietal cortex to 4-fold in the frontal cortex. FIGS. 17E and 17F show the signal distribution in k-space in the first and the last slice for a receiver coil located adjacent to the frontal cortex, which illustrates the increase in signal dispersion in k-space with measurement time. Box 1700 in FIG. 17F corresponds to the conventional k-space coverage without k-space expansion.

(59) The k-space expansion method can be expanded to 3D spatial encoding: The preferred implementation of k-space undersampling for a volumetric PEPSI acquisition employs radial undersampling in the k.sub.y-k.sub.z plane that increases with increasing radius.

(60) While the foregoing description applies to MR spectroscopic imaging, the k-space expansion methodology also applies to gradient echo MR imaging and functional MR imaging, reducing magnetic field inhomogeneity related signal losses and sensitivity to movement-related signal changes in regions with magnetic field inhomogeneity.

(61) While the foregoing written description enables one of ordinary skill to make and use what is considered presently to be the best mode thereof, those of ordinary skill will understand and appreciate the existence of variations, combinations, and equivalents of the specific embodiment, method, and examples herein. The disclosure should therefore not be limited by the above-described embodiments, methods, and examples, but by all embodiments and methods within the scope and spirit of the disclosure.