METHOD AND APPARATUS FOR RECONSTRUCTING MAGNETIC RESONANCE IMAGE DATA

0Associated
Cases 
0Associated
Defendants 
0Accused
Products 
0Forward
Citations 
0
Petitions 
4
Assignments
First Claim
1. A method for reconstructing magnetic resonance (MR) image data from kspace data, the method comprising:
 providing a computer with kspace data of an image region of a subject;
also providing the computer with a plurality of navigator signals for the image region of the subject;
in said computer, sorting the kspace data into a plurality of bins (103), the plurality of bins representing different motion states of the subject;
for each of the plurality of bins, executing a compressed sensing procedure in said computer to reconstruct the MR image data from the kspace data in the respective bin, execution of the compressed sensing procedure comprises solving an optimization problem comprising a data consistency component and a transform sparsity component; and
in said computer, incorporating motion information for the subject obtained from the plurality of navigator signals into at least one of the data consistency component and the transform sparsity component of the optimization problem.
4 Assignments
0 Petitions
Accused Products
Abstract
In a method for reconstructing magnetic resonance (MR) image data from kspace data, kspace data of an image region of a subject are provided to a computer that is also provided with multiple navigator signals for the image region of the subject. The computer sorts the kspace data into multiple bins, the multiple bins representing different motion states of the subject. For each of the multiple bins, the computer executes a compressed sensing procedure to reconstruct the MR image data from the kspace data in the respective bin. Execution of the compressed sensing procedure includes solving an optimization problem comprising a data consistency component and a transform sparsity component. Motion information is incorporated by the computer into at least one of the data consistency component and the transform sparsity component of the optimization problem.
0 Citations
No References
No References
16 Claims
 1. A method for reconstructing magnetic resonance (MR) image data from kspace data, the method comprising:
providing a computer with kspace data of an image region of a subject; also providing the computer with a plurality of navigator signals for the image region of the subject; in said computer, sorting the kspace data into a plurality of bins (103), the plurality of bins representing different motion states of the subject; for each of the plurality of bins, executing a compressed sensing procedure in said computer to reconstruct the MR image data from the kspace data in the respective bin, execution of the compressed sensing procedure comprises solving an optimization problem comprising a data consistency component and a transform sparsity component; and in said computer, incorporating motion information for the subject obtained from the plurality of navigator signals into at least one of the data consistency component and the transform sparsity component of the optimization problem.  View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13)
 14. A method for operating a magnetic resonance (MR) image apparatus comprising an MR data acquisition scanner, said method comprising:
operating the MR data acquisition scanner to acquire kspace data from an image region of a subject in the MR data acquisition scanner, and to obtain a plurality of navigator signals from said image region; providing the kspace data and the plurality of navigator signals to a computer and, in said computer, sorting the kspace data into a plurality of bins (103), the plurality of bins representing different motion states of the subject; for each of the plurality of bins, executing a compressed sensing procedure in said computer to reconstruct the MR image data from the kspace data in the respective bin, execution of the compressed sensing procedure comprises solving an optimization problem comprising a data consistency component and a transform sparsity component; and in said computer, incorporating motion information for the subject obtained from the plurality of navigator signals into at least one of the data consistency component and the transform sparsity component of the optimization problem.
 15. A magnetic resonance (MR) apparatus comprising:
an MR data acquisition scanner comprising a gradient system, and an excitation system; a computing system in communication with the excitation system, and the gradient system, and configured to receive the signals from the excitation system; the computing being configured to execute program code in order to control the gradient system so as to apply a magnetic field gradient in said scanner and in order to control the excitation system so as to apply an excitation pulse to a subject in the scanner and to receive signals from the subject resulting from the excitation that represent kspace data of an image region of a subject, and to acquire a plurality of navigator signals for the image region of the subject; the computing system being configured to sort the kspace data into a plurality of bins, the plurality of bins representing different motion states of the subject; the computing system being configured to, for each of the plurality of bins, execute a compressed sensing procedure to reconstruct the MR image data from the kspace data in the respective bin, execution of the compressed sensing procedure comprising solving an optimization problem comprising a data consistency component and a transform sparsity component, and the computing system is configured to incorporate motion information for the subject obtained from the plurality of navigator signals into at least one of the data consistency component and the transform sparsity component of the optimization problem.
 16. A nontransitory, computerreadable data storage medium encoded with programming instructions, said storage medium being loaded into a computer and said programming instructions causing said computer to:
receive kspace data obtained from an image region of a subject situated in an MR data acquisition scanner; also receive a plurality of navigator signals of the image region of the subject; sort the kspace data into a plurality of bins, the plurality of bins representing different motion states of the subject; for each of said plurality of bins, execute a compressed sensing procedure to reconstruct MR image data from the kspace data in the respective bin, the execution of the compressed sensing procedure comprising solving an optimization problem comprising a data consistency component and a transform sparsely component; and incorporate motion information for the subject obtained from the plurality of navigator signals into at least one of the data consistency component and the transform sparsity component of the optimization problem.
1 Specification
The present invention concerns a method of reconstructing magnetic resonance (MR) image data, and a nontransitory, computer readable data storage medium, and a MR imaging apparatus that implement such a method.
In particular the invention is concerned with a method for reconstructing magnetic resonance image data from kspace data, and in particular undersampled kspace data.
In magnetic resonance (MR) imaging, imaging data is commonly acquired as samples of the Fourier transform of the subject to be reconstructed. These samples are referred to as “kspace” data. Images are generated from the kspace data using an image reconstruction process which can involve recovering an estimate of the original subject from the kspace data.
Only a limited amount of kspace data may be acquired at any given time. There may also be delays between the kspace data acquisitions due to the time required for the signal excitation and encoding processes inherent to MR imaging. This typically means that the kspace data for an imaging region of the subject is acquired over several respiratory cycles of the subject. This may mean that respiratory motion artefacts are introduced into the kspace data, and as a result the quality of the reconstructed images may be reduced.
Respiratory motion is a particular problem in the freebreathing imaging of a volume region of a subject. One particular application of such is threedimensional wholeheart coronary MR angiography (CMRA).
One existing approach for minimizing respiratory motion artefacts is to use navigator pulses in the MR imaging sequence. The navigator pulses are typically applied so as to excite 1dimensional beams or 2dimensional planes of tissue. The navigator pulses result in a navigator signal (an echo signal) being returned, and which may be reconstructed in the readout direction. In particular, one existing approach uses navigator pulses to track the movement of the diaphragm of the subject, and acquires MR data using a gated approach. This means that MR data is only acquired when the detected motion of the diaphragm of the subject is within an acceptable range. Alternatively, the MR data may also be acquired outside the acceptable range, but such MR data may be rejected and reacquired.
This diaphragmatic respiratory motion approach may help to minimize ghosting and blurring artefacts caused by respiratory motion. However, this approach may only act to minimize the effect of motion in the superiorinferior (SI) direction, and may not account for motion components in other directions. Moreover, this existing approach has a low scan efficiency because MR data is only acquired when the subject motion is within an acceptable range or is acquired but rejected when outside the acceptable range and subsequently reacquired. This leads to prolonged scan times.
Another approach known as XDGRASP (eXtraDimensional Goldenangle Radial Sparse Parallel MR imaging) has been proposed to achieve approximately 100% scan efficiency, meaning that no data is rejected, while still compensating for respiratory induced motion of the subject.
XDGRASP uses a nonCartesian kspace sampling scheme, and in particular a radial kspace sampling scheme with a goldenangle (111.25°) separation between radial samples.
Radial kspace sampling means that the kspace samples may be designated as lying in the k_{x}k_{y }plane in kspace, and that adjacent kspace samples may be angularly separated from one another such that each kspace sample represents a spoke of a wheel. Goldenangle separation means that there is a 111.25° angular increment between consecutive spokes. Radial kspace sampling contrasts with Cartesian kspace sampling where the kspace samples are rectilinear. That is, the kspace samples run parallel to one another.
Due to the nature of the radial sampling scheme, the center of kspace is sampled repeatedly. This means that radial sampling has reduced sensitivity to kspace gaps due to motionrelated phase shifts. As a result, the XDGRASP sequence helps minimize the effect of motion on the reconstruction.
XDGRASP may be used to obtain kspace data for a 2D image region or a 3D image volume. In order to generate kspace data for a 3D image volume, XDGRASP uses a 3D stackofstars sampling scheme, in which golden angle radial sampling is used in the k_{x}k_{y }plane and Cartesian sampling is used in the k_{z }dimension. Another XDGRASP approach uses a 3D spiral phyllotaxis trajectory.
XDGRASP sorts the acquired kspace data into multiple sets of undersampled datasets with distinct motion states using motion signals extracted directly from the kspace data.
In particular, the center of kspace (k_{x}=k_{y}=0 for 2D and k_{x}=k_{y}=k_{z}=0 for 3D) in each projection angle can be used to obtain the temporal variation caused by subject motion. In addition or alternatively, for 3D, a 1D projection may be used for selfnavigation when a spiral phyllotaxis trajectory is used. The center of kspace corresponds to a constant term in the Fourier representation of the image, whose magnitude is the same as the average image brightness. Thus, by comparing the center of kspace values for the different projection angles, the change in the center of kspace value across the different projection angles can be determined, and used to estimate the subject respiratory motion.
As a result of the sorting process used by XDGRASP, the acquired data is sorted into multiple respiratory phases, which are known as and will subsequently be referred to as respiratory bins. The respiratory bins of kspace data are then reconstructed to generate respiratory resolved images. This reconstruction is performed using a compressed sensing algorithm which exploits total variation (TV) sparsity along the respiratory dimension.
In particular, XDGRASP reconstructs each undersampled respiratory bin using a compressed sensing procedure. In particular, XDGRASP solves an optimization problem of the form:
Here, x is the respiratoryresolved image series, k is the binned kspace data. The encoding operator E=AFS incorporates the sampling matrix A for each bin i, Fourier transform F and coil sensitivities S.
Here, ∥ ∥_{2}^{2 }denotes the l_{2}norm defined as ∥u∥_{2}^{2}=√(Σ_{i}u_{i}^{2}), for a given vector u.
Here, Ψ_{t }is a 1D temporal total variation (TV) function applied in the respiratoryresolved direction. Here, α is a regularization parameter.
The component ½∥Ex−k∥_{2}^{2 }of the optimization problem is known as a data consistency component of the compressed sensing procedure.
The component Ψ_{t}(x) of the optimization problem is known as a transform sparsity component of the compressed sensing procedure.
The parameter α is a regularization parameter that determines the tradeoff between the data consistency and the sparsity.
While XDGRASP has been successful in providing respiratory resolved 3D radial CMRA images with 100% scan efficiency, XDGRASP has been shown to suffer from reduced image quality when using a Cartesian rather than a radial sampling scheme. A Cartesian sampling scheme may be used in many applications including, for example, Cartesian liver imaging. The reduced image quality is particularly pronounced for respiratory phases with large respiratory displacements.
An object of the present invention is to provide an improved approach for reconstructing MR image data from kspace data, or at least provide an alternative to the existing reconstruction approaches. It is a particular objective to provide an improved reconstruction approach that is robust to motion even when a Cartesian sampling scheme is used to acquire the kspace data.
According to a first aspect of the invention, there is provided a method of reconstructing MR image data from kspace data. The method includes providing a computer with kspace data of an image region of a subject and providing the computer with a number of navigator signals for the image region of the subject. The method includes sorting the kspace data into a number of bins, the number of bins representing different motion states of the subject. The method further includes, for each of the multiple bins, executing a compressed sensing procedure in the computer to reconstruct the MR image data from the kspace data in the respective bin. The compressed sensing procedure includes solving an optimization problem involving a data consistency component and a transform sparsity component. Motion information for the subject obtained from the number of navigator signals is incorporated by the computer into at least one of the data consistency component and the transform sparsity component of the optimization problem.
Significantly, the present invention acquires navigator signals for the image region of the subject, and uses motion information of the subject obtained from the plurality of navigator signals to reconstruct the bins of kspace data as part of a compressed sensing procedure. In this way, the motion information may be used to provide robustness against subject motion in the reconstruction even if a Cartesian sampling scheme is used. Further, the present invention is not required to acquire the kspace data in a motion gated way because the motion information is used in the reconstruction. As such, kspace data may be acquired with close to or even 100% scan efficiency.
Generally, the present invention uses motion information to increase sparsity. The motion information may relate to subject motion such as respiratory motion or cardiac motion. In this way, the present invention is able to minimize respiratory motion artefacts and artefacts due to cardiac motion. As such, respiratory and/or cardiac resolved MR image data may be reconstructed.
The acquired kspace data may be subsampled (undersampled) kspace data. By subsampled, we mean that the kspace data is acquired using fewer echoes than a conventional MR scan. In other words, only data for part of the kspace is collected. The kspace data may be acquired using an incoherent sampling scheme. This may be so as to create a noiselike appearance for the aliasing artefacts that are generated as a result of the subsampling. The noiselike artefacts may be removed in a subsequent step of the optimization procedure.
As explained above, the MR image data are reconstructed from the kspace data using a compressed sensing procedure. The compressed sensing procedure comprises solving an optimization problem that comprises a data consistency component and a transform sparsity component.
The data consistency component acts to ensure that the reconstructed MR image data is consistent with the original acquired kspace data. That is, the data consistency component may be a constraint on the optimization problem that takes into account the difference between the MR image data transformed into kspace and the original acquired kspace data. The optimization problem may seek to minimize the difference between the MR image data transformed into kspace and the original acquired kspace data. This minimization may be subject to other constraints, e.g. the transform sparsity component.
The transform sparsity component of the compressed sensing procedure acts to increase the sparsity of the MR image data in a sparse representation space known as Wspace. The increase in sparsity corresponds to a diminishing of the aliasing artefacts in the image space. The optimization problem may seek to maximize the sparsity. This maximization may be subject to other constraints, e.g. the data consistency component.
Thus, the compressed sensing procedure refers to a procedure in which MR image data is reconstructed that is both consistent with the original acquired kspace data and also forms a sparse representation when transformed into Wspace.
The kspace data may be acquired over a number of cardiac cycles for the subject. Each of the number of navigator signals may be acquired during a respective one of the number of cardiac cycles.
The number of bins may represent different motion states of the subject. The different motion states may be different respiratory phases of the subject.
Sorting the kspace data into the number of bins may sort the kspace data into the number of bins using the motion information for the subject obtained from the plurality of navigator signals.
The method may further include correcting for motion within the number of bins using the motion information for the subject obtained from the number of navigator signals so as to generate motion corrected binned kspace data. The data consistency component may use the motion corrected binned kspace data. In other words, intrabin motion correction may be performed. Significantly, correcting for motion within the plurality of bins helps to improve the image quality of each bin, and thus minimize the effect of motion on the reconstruction.
Correcting for motion within the plurality of bins may involve correcting for translational motion within the number of bins. The translational motion may be 2D translational motion. The motion information may be beattobeat translation information. The beattobeat translation information may be 2D beattobeat translation information.
The data consistency component of the optimization problem may be a determination of the difference between the reconstructed MR image data, x, and the motion corrected binned kspace data, b.
The data consistency component may be a determination of the difference between the reconstructed MR image data, x, transformed into kspace and the motion corrected binned kspace data, b.
The determination of the difference may be determining the l_{2}norm of the result of a subtraction of the motion corrected binned kspace data from the reconstructed MR image data transformed into kspace.
The data consistency component may be the function ½∥Ex−b∥_{2}^{2},
wherein x is the reconstructed MR image data,
wherein b is the motion corrected binned kspace data,
wherein E is an encoding operator for transforming the reconstructed MR image data into kspace, and
wherein ∥ ∥_{2}^{2 }denotes the l_{2}norm defined as ∥u∥_{2}^{2}=√(Σ_{i}u_{i}^{2}), for a given vector u.
The reconstructed MR image data x, may be of the form: x=[x_{1}, x_{2}, . . . , x_{N}]^{T}, wherein N is the number of bins. It will be appreciated that optimal values for x are desired to be determined by solving the optimization problem.
The motion corrected binned kspace data b, may be of the form: b=[b_{1}, b_{2}, . . . , b_{N}]^{T}, wherein N is the number of bins.
Significantly, unlike the existing XDGRASP approach which incorporates the function ½∥Ex−k∥_{2}^{2 }in the reconstruction problem, where k is the binned kspace data (i.e. not motion corrected), the present invention may incorporate motion corrected binned kspace data, b, in the reconstruction problem.
The encoding operator may be defined as E=AFS. The encoding operator may thus incorporate the sampling matrix A for each bin b, Fourier transform F and coil sensitivities S.
The transform sparsity component may comprise a sparsifying transform Ψ_{t }for transforming the reconstructed MR image data, x, into Wspace. That is, the sparsifying transform is an operator that maps the MR image data into a sparse Wspace.
The sparsifying transform may be a temporal (e.g. onedimensional) total variation function. The total variation function may be applied in the forward direction, e.g. the respiratory resolved direction.
The sparsifying transform may be a wavelet transform. Other forms of sparsifying transform are within the scope of the present invention.
The reconstructed MR image data, x, may first be transformed to a motioncorrected domain using the motion information for the subject obtained from the plurality of navigator signals. The motioncorrected reconstructed MR image data may then undergo the sparsifying transform. In other words, the transform sparsity component transforms the reconstructed MR image data to a motioncorrected domain using the motion information for the subject obtained from the plurality of navigator signals so as to generate motioncorrected reconstructed MR image data, and applies a sparsifying transform to the motioncorrected reconstructed MR image data.
Significantly, incorporating the motion information for the subject into the sparsifying transform helps increase the sparsity within the motion domain of the subject (e.g. the respiratory domain). This is because the MR image data is aligned to one motion position (e.g. respiratory position) during the reconstruction.
The sparsifying transform may be the function Ψ_{t }(x) where x is the motioncorrected domain.
x may be a motioncorrected domain and may be determined such that x=T_{b}x_{b}, wherein T_{b }is a translation transform that maps a bin image x_{b }to a reference image x_{r }determined using the motion information. The reference image may be an endexpiration image.
The sparsifying transform component may comprise performing a finite differences operation on the motioncorrected reconstructed MR image data, x.
The transform sparsity component may comprise determining the l_{1}norm of the result of the finite differences operation performed on the motioncorrected reconstructed MR image data.
In one example, the sparsifying transform may be defined as:
In this example, ∇_{t }is a finite differences operator along the temporal motion (e.g. respiratory) dimension, I is the identity matrix and is the translation transform that maps the bin image x_{b }to the reference image x_{1 }(endexpiration).
In this example, =I. But this is not required in all examples.
∥u∥_{1 }is the l_{1}norm and is defined as ∥u∥_{1}=Σ_{i=1}^{n}u(i), where i is the ith element of u and n is the total number of elements.
The transform sparsity component may be a sparsifying transform in the forward direction and a sparsifying transform in the backward direction. The sparsifying transform in the forward direction may be of the form Ψ_{t }(x). The sparsifying transform in the backward direction may be of the form Ψ_{−t}(x).
In one example, the optimization problem may thus be a transform sparsity component of the form: αΨ_{t}(x)+γΨ_{−t}(x), wherein γ is a regularization parameter.
The use of a sparsifying transform in the forward and backward directions may help exploit temporal redundancies in both forward and backward directions. This may help to provide a sparser representation of the motioncorrected domain. In particular, finite differences sparsity along the motion (e.g. respiratory) dimension may be calculated between every motion (e.g. respiratory) phase, x_{i}, and its two adjacent motion phases, x_{i−1}, and x_{i+1}, in order to exploit the temporal redundancies in both forward and backward directions. This approach may also involve applying a translation transformation to map x_{i−1 }to x_{i }and another transformation to map x_{i+1 }to x_{i}.
The optimization problem may further comprise a spatial total variation function of the form βΨ_{s}(x), wherein β is a regularization parameter, and wherein Ψ_{s }(x) is a spatial total variation function. The spatial total variation function may be a threedimensional spatial total variation function.
Both the data consistency component and the transform sparsity component of the optimization problem may incorporate the motion information for the subject obtained from the plurality of navigator signals.
The optimization problem may be of the form
wherein ƒ(x) is a function defined as {data consistency component+transform sparsity component}. The optimization problem may thus comprise determining a value for the reconstructed MR image data set, x, at which a function ƒ(x) attains its minimum.
The optimization problem may be of the form:
Here, α is a regularization parameter that defines the tradeoff between the data consistency and the sparsity.
The optimization problem may be of the form:
Here, Ψ_{t }(x) is a sparsifying function.
In equation (2) the optimization problem incorporates the motion information in the data consistency component, but does not incorporate motion information in the transform sparsity component.
The optimization problem may be of the form:
In equation (3) the optimization problem incorporates the motion information in the transform sparsity component, but does not incorporate motion information in the data consistency component.
The optimization problem may be of the form:
In equation (4) the optimization problem incorporates the motion information in the transform sparsity component and the data consistency component.
The optimization problem may be of the form:
The optimization problem may be of the form:
The optimization problem may be of the form:
The above optimization problems may further comprise a sparsifying transform in the backward direction. This sparsifying transform may be of the form Ψ_{−t}(x) or Ψ_{−t}(x).
The compressed sensing procedure may be a first reconstruction procedure, and the method may further include performing a second reconstruction procedure using the MR image data reconstructed by the first reconstruction procedure.
Performing the second reconstruction procedure includes using the MR image data reconstructed using the first reconstruction procedure to determine one or more nonrigid motion fields, and, for each of the number of bins, using the one or more nonrigid motion fields to reconstruct the MR image data from the kspace data in the respective bin. The nonrigid motion fields are estimated using a nonrigid image registration algorithm. The nonrigid motion fields may describe the motion and deformation of the heart. They may be displacement vectors defined at each voxel that find the correspondence between a bin image and the reference image, in order to align the images. These not rigid motion fields may also be used in equation (7) or other similar equations as a replacement for the operator . This would involve solving the optimization twice. Since this is a more accurate image registration, performed on a 3D high resolution image, the sparsity should increase. The algorithm may be based on freeform deformations that are modelled by bsplines. This produces a set of nonrigid deformations fields, relating each motion phase (e.g. respiratory phase) to a reference state (e.g. endexpiration).
Performing the second reconstruction of the MR image data may also include solving an optimization problem comprising a data consistency component.
The data consistency component may be an operator, G, that incorporates information from nonrigid motion fields estimated from the reconstructed MR image data.
The optimization procedure for the second reconstruction seeks to minimize the difference between the reconstructed MR image data, x, as transformed into kspace using the operator, G, and the acquired kspace data, k or the motion corrected kspace data, b.
The data consistency component seeks to minimize the l_{2}norm of the difference between the reconstructed MR image data, x, as transformed into kspace using the operator, G, and the acquired kspace data, k or the motion corrected kspace data, b.
The operator G may be determined as G=Σ_{i}E_{i}U_{i},
wherein E is an encoding operator for the bin i, and
wherein U is the one or more nonrigid motion fields estimated from the MR image data reconstructed using the first reconstruction procedure for the bin i.
The encoding operator may be defined as E=AFS. The encoding operator may thus incorporate the sampling matrix A for each bin i, Fourier transform F and coil sensitivities S.
The optimization problem for the second reconstruction may be of the form:
wherein G is as defined above, and where k is the binned kspace data.
The optimization problem for the second reconstruction may be of the form:
wherein G is as defined above, and where b is the motion corrected binned kspace data.
The optimization problem for the second reconstruction may further comprise a spatial total variation function of the form βΨ_{s}(x), wherein β is a regularization parameter, and wherein Ψ_{s }(x) is a spatial total variation function. The spatial total variation function may be a threedimensional spatial total variation function.
The nonrigid motion fields used in the second optimization procedure may be used instead in a subsequent first optimization procedure so as to further increase the sparsity in the sparsifying transform Ψ_{t }(x) instead of or in addition to the motion information. That is, the nonrigid motion fields can be sued to generate respiratory or other motion resolved nonrigid motioncompensated images.
The kspace data may be acquired using a nonCartesian sampling scheme, which may be a radial sampling scheme. The kspace data may be acquired using a radial sampling scheme with golden angle separation.
The kspace data may be acquired using a Cartesian sampling scheme. The kspace data may be acquired using a goldenstep Cartesian trajectory with spiral or radial profile ordering. Significantly, existing approaches such as XDGRASP generally use a radial sampling scheme due to the increased robustness to motion provided by such radial sampling schemes. This is because, as explained previously, radial sampling schemes densely sample the central region of kspace and the center of kspace, itself, is sampled multiple times. Cartesian sampling does not sample the center of kspace repeatedly, and thus does not provide the same robustness to motion as radial sampling. However, Cartesian sampling offers benefits over radial sampling in many applications. These benefits include a higher signaltonoise ratio efficiency, i.e. per unit of measurement time, lower sensitivity to offresonance and gradient delays, and reduced reconstruction complexity.
The present invention is able to achieve the benefits of Cartesian sampling while mitigating/reducing the negatives of Cartesian sampling by using the motion information for the subject obtained from the plurality of navigator signals in the reconstruction of the MR image data. The use of motion information helps compensate for motion in the kspace data and thus reduce the susceptibility of the reconstruction to motion even if a Cartesian sampling scheme is used. Respiratory and/or cardiac resolved MR image data may thus be reconstructed.
The navigator signals may be generated from image navigator pulses. The image navigator pulses may be 1D, 2D, or 3D image navigator pulses.
The use of 3D image navigator pulses may be beneficial in certain applications to estimate beattobeat 3D translational motion, and thus, additionally allowing for motion correction in the anteriorposterior direction.
The kspace data may be for a 2D image region of the subject, or may be for a 3D image region of the subject.
The navigator signals may be 1D, 2D, or 3D navigator signals.
According to a second aspect of the invention, there is provided a MR apparatus that includes a computing system, and an MR data acquisition scanner having a basic field magnet, a gradient system, and an excitation system. The gradient system is operated to apply a magnetic field gradient. The excitation system is operated to apply an excitation pulse to the subject and to receive signals from the subject. The computing system is in communication with the excitation system, and the gradient system for controlling these components. The computing system receives the signals from the excitation system. The computing system is further configured to execute program code to control the gradient system and the excitation system to acquire kspace data of an image region of a subject, and a number of navigator signals for the image region of the subject. The computing system is further configured to sort the kspace data into a number of bins, the number of bins representing different motion states of the subject. The computing system is further configured to, for each of the number of bins, execute a compressed sensing procedure to reconstruct the MR image data from the kspace data in the respective bin. The compressed sensing procedure includes solving an optimization problem involving a data consistency component and a transform sparsity component. Motion information for the subject obtained from the number of navigator signals is incorporated by the computing system into at least one of the data consistency component and the transform sparsity component of the optimization problem.
The computing system is configured to execute program code for controlling the MR apparatus to perform the method as described above in relation to the first aspect.
According to a third aspect of the invention, there is provided a nontransitory computer readable data storage medium encoded with programming instructions, when executed by a computing device, cause the computing device to perform the method as described above in relation to the first aspect.
The present invention relates to compressed sensing procedures which comprise solving optimization problems comprising a data consistency component and a transform sparsity component.
An example of a compressed sensing procedure suitable for use with the present invention will now be described which includes the following steps (1) to (9).
In step (1) of this example of the inventive compressed sensing procedure, the original acquired kspace data is transformed into MR image data (e.g. a MR image), e.g. through use of a Fourier transform. The MR image data may suffer from subsampling artefacts. However, if an incoherent sampling scheme is used to acquire the kspace, these artefacts may appear to be ‘smeared’ over the image. That is, almost homogeneous noiselike artefacts may appear in the MR image data. The reconstructed MR image data may serve as a starting point for the optimization problem. The optimization problem seeks to find a better solution (a better set of MR image data) that is also consistent with the original acquired kspace data.
In step (2) of this example, the MR image data is transformed into a sparse representation (Wspace).
The purpose of the sparse transformation is to attempt to locally separate the desired signal from the noise artefacts. The Wspace is a better suited depiction of the image as the sparsity in Wspace is higher than in the original image space. A higher sparsity means that the useful image content is concentrated in a few pixels in Wspace, while most of the pixels have only a very low signal.
After the transformation into Wspace, the useful image content is better separated from the noise artefacts.
In step (3), the Wspace is thresholded to remove noise.
One example thresholding procedure which may be used is referred to as ‘soft thresholding’ in which all pixels with a value less than the threshold are set to zero, and then the threshold is subtracted from all other pixels. As a result of the thresholding procedure, many pixels in the Wspace have a value of 0, and as such there are fewer nonzero pixels. Consequently, the sparsity of the image in Wspace is increased.
In step (4), the Wspace is transformed back into the MR image data.
The Wspace representation may be transformed back into the image space using an inverse W transformation. The transformed MR image data has less noise due to the thresholding procedure described above, but does not exactly reflect the original acquired kspace data. As a result, the compressed sensing procedure may then check the image consistency. That is, how well the denoised MR image data still represents the acquired kspace data.
In step (5), the MR image data of step (4) is transformed into kspace. An inverse Fourier transform may be applied to transform the MR image data of step (4) back into kspace. It will be appreciated that the kspace data will cover all spatial frequencies while the original subsampled acquired kspace data covers only a portion of the frequencies.
In step (6) the kspace data of step (5) is filtered. The kspace data of step (5) may be filtered such that the filtered kspace data only depicts the spatial frequencies that were acquired during the original kspace acquisition. The filtering may comprise setting all of the kspace data of step (5) to zero except for the points of kspace data that correspond to the spatial frequencies acquired during the original kspace data acquisition.
In step (7) the kspace data of step (6) is subtracted from the original kspace data. The difference between the kspace data of step (6) and the original kspace data may correspond to the error that the thresholding of step (3) has created compared to the original acquired kspace.
In step (8), the ‘difference’ kspace data of step (7) is transformed into the MR image space so as to form a ‘difference’ MR image. The kspace data may be transformed using a Fourier transformation.
In step (9), the MR image data of step (1) is updated with the ‘difference’ MR image of step (9).
The updating may comprise adding the ‘difference’ MR image of step (9) to the MR image data of step (1). The updated MR image data may have less noiselike artefacts (corresponding to a higher sparsity in Wspace) than the image had before the update. At the same time, the updated MR image data has been made consistent with the measured kspace.
Steps (3)(9) may be repeated, e.g. as part of an iterative optimization procedure. Each iteration may increase the sparsity (in Wspace), which corresponds to diminishing the aliasing artefacts in image space. At the same time, the consistency of the reconstruction with measured kspace is taken care of.
The following examples may use the compressed sensing procedure of the form as described generally above. It will further be appreciated that the present invention is not directed towards a particular compressed sensing procedure, but rather the beneficial incorporation of motion information for the subject obtained from the plurality of navigator signals into at least one of the data consistency component and the transform sparsity component of the optimization problem of the compressed sensing procedure.
Referring to
Step 101 concerns obtaining kspace data of an image region of a subject.
In the example of
The present invention is not limited to the particular kspace acquisition procedure of the example of
Step 102 concerns obtaining a number of navigator signals for the image region of the subject.
In the example of
The present invention is not limited to the particular navigator signal acquisition procedure of the example of
Step 103 concerns sorting the kspace data into a number of bins, the number of bins representing different motion states of the subject.
In the example of
The present invention is not limited to the sorting procedure of
Step 104 includes, for each of the number of bins, executing a compressed sensing procedure to reconstruct the MR image data from the kspace data in the respective bin, the compressed sensing procedure comprises solving an optimization problem comprising a data consistency component and a transform sparsity component, wherein motion information for the subject obtained from the number of navigator signals is incorporated into at least one of the data consistency component and the transform sparsity component of the optimization problem.
In the example of
In particular, step 104 includes correcting for motion within the plurality of bins using the motion information for the subject obtained from the number of navigator signals so as to generate motion corrected binned kspace data. The correcting for motion within the number of bins involves correcting for translational motion within the number of bins. In this example, the translational motion is 2D translational motion, and in particular is beattobeat translation information. Or, in other words, the translational motion of the subject between heartbeats. The beattobeat translation information is 2D beattobeat translation information, e.g. obtained from 2D image navigator signals. This means, in this example, that 2D translational motion correction within each bin is performed in kspace before the reconstruction to improve image quality of each bin.
In other words, in the example of
The present invention is not limited to the correcting procedure of
In the example of
In particular, the data consistency component of the optimization problem comprises the function ½∥Ex−b_{2}^{2},
wherein x is the reconstructed MR image data,
wherein b is the motion corrected binned kspace data,
wherein E is an encoding operator, and
wherein ∥ ∥_{2}^{2 }denotes the l_{2}norm defined as x_{2}^{2}=√(Σ_{i}x_{i}^{2}).
In one example implementation, the optimization problem may be of the form:
Ψ_{t }(x) is a sparsifying transform and α is a regularisation parameter.
In another example implementation, the optimization problem may further incorporate the motion information for the subject obtained from the plurality of navigator signals in the transform sparsity component of the optimization problem. In this example, the transform sparsity component comprises a temporal total variation function. The motion information for the subject obtained from the plurality of navigator signals is then incorporated into the temporal total variation function. This means that the images are aligned to one motion state (e.g. respiratory position) to increase sparsity in the motion (e.g. respiratory) domain.
In this example, the sparsifying transform is the function Ψ_{t }(x) where x is the motioncorrected domain.
x is a motioncorrected domain and may be determined such that x=T_{b}x_{b}, wherein T_{b }is a translation transform that maps a bin image x_{b }to a reference image x_{r }determined using the motion information. The reference image may be an endexpiration image.
In this particular example implementation, the sparsifying transform Ψ_{t }(x), is defined as:
In this example, ∇_{t }is a finite differences operator along the temporal motion (e.g. respiratory) dimension, I is the identity matrix and is the translation transform that maps the bin image x_{b }to the reference image x_{1 }(endexpiration).
In this example, =I.
∥u∥_{1 }is the l_{1}norm and is defined as ∥u∥_{1}=Σ_{i=1}^{n}u(i), where i is the ith element of u and n is the total number of elements.
In this example implementation, the optimization problem may be of the form:
In another example implementation, the optimization problem further comprises a spatial total variation function of the form βΨ_{s}(x), wherein β is a regularization parameter, and wherein Ψ_{s}(x) is a spatial total variation function. The spatial total variation function may be a threedimensional total variation function.
In this example implementation, the optimization problem may be of the form:
In this or another example implementation, the optimization problem may only incorporate the motion information in the data consistency component of the optimization problem. The optimization problem may be of the form:
As discussed above, the present invention does not require that the motion within the number of bins is corrected for in all embodiments. In other embodiments, the motion information for the subject obtained from the number of navigator signals may only be incorporated in the transform sparsity component of the optimization problem.
In one example implementation, the optimization problem may be of the form:
Here, k is the binned kspace data, i.e. kspace that is not motion corrected.
In another example implementation, the optimization problem may be of the form:
The above optimization problems may further comprise a sparsifying transform in the backward direction. This sparsifying transform may be of the form Ψ_{−t}(x) or Ψ_{−t}(x).
By additionally using a sparsifying transform in the backward direction, the sparsity is able to be calculated between two or more adjacent respiratory phases because the respiratory motion is approximately periodic. This approach may provide a sparser representation in the motioncorrected respiratory domain, particularly because this approach depends less on the quality of the endexpiration image and the accuracy of the motion model used in the reconstruction.
The optimization problems mentioned above can all be solved using nonlinear conjugate gradient methods. Other methods for solving compressed sensing optimization problems are within the scope of the present invention.
The reconstruction methods described above have been found to produce highquality respiratoryresolved images for all respiratory phases, allowing clear visualization of coronary arteries in CMRA. The reconstruction method helps to reduce respiratory motion, and thus, improves the visualized vessel length and sharpness of both coronaries compared to the existing Cartesian XDGRASP approach.
Incorporating the motion information into the data consistency component of the optimization problem, e.g. by performing intrabin translation motion correction, has been found to improve the quality of the respiratory resolved images. Additionally or separately, incorporating the motion information into the transform sparsity component of the optimization problem has been found to increase the sparsity in the respiratory dimension, and hence improve the quality of the image reconstruction. Further, incorporating the spatial total variation function into the transform sparsity component has been found to produce a denoising effect in the resulting image reconstruction, and thus also improves the quality of the image reconstruction. In addition, the spatial total variation function may also help handling undersampled data.
In example implementations, the optimization problems described above are a first stage of the compressing sensing reconstruction procedure.
In these example implementations, the method further comprises performing a second reconstruction procedure using the MR image data reconstructed using the first reconstruction procedure
In an example implementation, the second reconstruction procedure comprises using the MR image data reconstructed using the first reconstruction procedure to determine one or more nonrigid motion fields. For each of the plurality of bins, the one or more nonrigid motion fields may be used to reconstruct the MR image data from the kspace data in the respective bin.
In particular examples, performing the second reconstruction of the MR image data comprises solving an optimization problem comprising a data consistency component. The data consistency component may comprise an operator G=Σ_{b }E_{b}U_{b},
wherein E is an encoding operator for the bin b, and
wherein U is the one or more nonrigid motion fields estimated from the MR image data reconstructed using the first reconstruction procedure for the bin b.
In these examples, the encoding operator is defined as E=AFS. The encoding operator thus incorporates the sampling matrix A for each bin b, Fourier transform F and coil sensitivities S.
In one example implementation, the optimization problem for the second reconstruction is of the form:
wherein G is as defined above, and where k is the binned kspace data.
In another example implementation, the optimization problem for the second reconstruction is of the form:
wherein G is as defined above, and where b is the motion corrected binned kspace data.
In the above examples, the number of the plurality of bins may be predetermined or selected by an operator, e.g. so as to provide an acceptable compromise between remaining intrabin motion and under sampling artefacts. In addition, the number of the plurality of bins may be selected according to the regularity or irregularity of the subject'"'"'s breathing pattern. For example, when the respiratory motion is relatively large, more respiratory phases could be added until all the bin widths are smaller than a certain value (e.g. 3 mm).
In some examples, the method may further comprise applying an autofocus technique to at least one or all of the bins. This may be applied to the bins to estimate, iteratively, the 3D translation parameters that maximize the sharpness of an image region of the subject (e.g. the coronary arteries). This approach may use estimated SI and LR motion as prior information, and further improve the bin images.
In some examples, intrabin residual translation motion may be addressed by acquiring 3D image navigator signals. 3D image navigator signals may be used to estimate beattobeat 3D translation motion, and thus additionally allow for motion correction in the anteriorposterior direction. Subsequently, these bin images could be used to estimate 3D translational or nonrigid interbin motion parameters at each iteration of the optimization procedure. These approaches may help further increase the sparsity in the motion (e.g. respiratory) dimension, and hence improve the quality of the respiratoryresolved reconstructions. In other approaches, the reconstructed MR data from the first reconstruction procedure may be used to estimate 3D bintobin nonrigid motion, which can be used to provide a singlephase 3D nonrigid motioncompensated image. As mentioned above, the nonrigid motion fields may also be used in the first optimization problem instead of the translational motion. For example, in equation (7) may be set as the 2D translational motion or the 3D nonrigid motion
The gradient system 203 is operated to apply a magnetic field gradient. The gradient system 203 is operated to apply magnetic field gradients along three spatial axes.
The excitation system 205 may include a transmitter (not shown) and a receiver (not shown). The excitation system 205 can be an RF system with one or more RF coils (not shown). The excitation system 205 is designed to apply an excitation pulse to the subject and to receive signals from the subject.
As noted, the scanner of the MR apparatus 200 includes a magnet (not shown) for establishing a stationary magnetic field. The magnet can include a permanent magnet, a superconducting magnet or other type of magnet.
The computing system 201 is in communication with the excitation system 205, and the gradient system 203 for controlling these components. The computing system 201 is configured to receive the signals from the excitation system 205.
The computing system 201 is further configured to execute program code to control the gradient system 203 and the excitation system 205 to acquire kspace data of an image region of a subject, and a number of navigator signals for the image region of the subject.
The computing system 201 is further configured to sort the kspace data into a number of bins, the number of bins representing different motion states of the subject.
The computing system 201 is further configured to, for each of the number of bins, execute a compressed sensing procedure (algorithm) to reconstruct the MR image data from the kspace data in the respective bin. The compressed sensing procedure involves solving an optimization problem comprising a data consistency component and a transform sparsity component. Motion information obtained from the number of navigator signals is incorporated into at least one of the data consistency component and the transform sparsity component of the optimization problem.
The computing system 201 is configured to execute program code for controlling the MR apparatus 200 to perform the method of the first aspect. The computing system 201 could be an integrated component of the MR apparatus 200. The computing system 201 could be a unit of a desktop computer, a workstation, a server, or a laptop computer.
According to aspects of the invention, there is also provided a computerreadable medium having instructions recorded thereon which, when executed by a processing unit, cause the processing unit to perform the method of the first or second aspect.
In other words, there is provided a method for operating an MR apparatus in order to reconstruct magnetic resonance (MR) image data from kspace data. The method includes obtaining kspace data of an image region of a subject (101). The method includes obtaining a number of navigator signals for the image region of the subject (102). The method includes sorting the kspace data into a number of bins, the plurality of bins representing different motion states of the subject (103). The method includes, for each of the number of bins, executing a compressed sensing procedure to reconstruct the MR image data from the kspace data in the respective bin (104). The compressed sensing procedure involves solving an optimization problem having a data consistency component and a transform sparsity component. Motion information obtained from the plurality of navigator signals is incorporated into at least one of the data consistency component and the transform sparsity component of the optimization problem.
At least some of the example embodiments described herein may be constructed, partially or wholly, using dedicated specialpurpose hardware. Terms such as ‘component’, ‘module’ or ‘unit’ used herein may include, but are not limited to, a hardware device, such as circuitry in the form of discrete or integrated components, a Field Programmable Gate Array (FPGA) or Application Specific Integrated Circuit (ASIC), which performs certain tasks or provides the associated functionality. In some embodiments, the described elements may be configured to reside on a tangible, persistent, addressable storage medium and may be configured to execute on one or more processors. These functional elements may in some embodiments include, by way of example, components, such as software components, objectoriented software components, class components and task components, processes, functions, attributes, procedures, subroutines, segments of program code, drivers, firmware, microcode, circuitry, data, databases, data structures, tables, arrays, and variables. Although the example embodiments have been described with reference to the components, modules and units discussed herein, such functional elements may be combined into fewer elements or separated into additional elements.
The described and illustrated embodiments are to be considered as illustrative and not restrictive in character, it being understood that only the preferred embodiments have been shown and described and that all changes and modifications that come within the scope of the inventions as defined in the claims are desired to be protected. It should be understood that while the use of words such as “preferable”, “preferably”, “preferred” or “more preferred” in the description suggest that a feature so described may be desirable, it may nevertheless not be necessary and embodiments lacking such a feature may be contemplated as within the scope of the invention as defined in the appended claims. In relation to the claims, it is intended that when words such as “a,” “an,” “at least one,” or “at least one portion” are used to preface a feature there is no intention to limit the claim to only one such feature unless specifically stated to the contrary in the claim. When the language “at least a portion” and/or “a portion” is used the item can include a portion and/or the entire item unless specifically stated to the contrary.
Although modifications and changes may be suggested by those skilled in the art, it is the intention of the Applicant to embody within the patent warranted hereon all changes and modifications as reasonably and properly come within the scope of the Applicant'"'"'s contribution to the art.