Method of reconstructing magnetic resonance image data
A plurality of sets of k-space data each of the same image region of a subject but having different contrasts are obtained. A sparse image coding procedure is performed to reconstruct a plurality of MR images each corresponding to one of the sets of k-space data. This involves solving an optimization problem comprising a data consistency iteration step used to generate the reconstructed MR images; and a denoising iteration step applied to the reconstructed MR images generated during the data consistency iteration step. The denoising iteration step includes performing a 2D/3D block matching operation to identify similar patches across the reconstructed MR images, and using the similar patches across the reconstructed MR images in a sparsifying operation to provide sparse representations of the reconstructed MR images. The sparse representations are used as an input to the data consistency iteration step.
1. A method of reconstructing magnetic resonance (MR) images from k-space data, the method comprising:
obtaining a plurality of sets of k-space data each of the same image region of a subject, wherein the sets of k-space data are acquired with different contrasts;
using a sparse image coding procedure to reconstruct a plurality of MR images, each corresponding to one of the sets of k-space data by solving an optimization problem, the optimization problem comprising:
a data consistency iteration step used to generate the reconstructed MR images, wherein each of the reconstructed MR images corresponds to one of the sets of k-space data;
a denoising iteration step applied to the reconstructed MR images generated during the data consistency iteration step, wherein the denoising iteration step comprises:
performing a 2D/3D block matching operation to identify similar patches across the reconstructed MR images; and
using the similar patches across the reconstructed MR images in a sparsifying operation to provide sparse representations of the reconstructed MR images, wherein the sparse representations are used as an input to the data consistency iteration step of the optimization problem.
2. A method as claimed in claim 1 , wherein the data consistency iteration step comprises determining a difference between the reconstructed plurality of MR images transformed into k-space and the acquired plurality of sets of k-space data.
3. A method as claimed in claim 2 , wherein the reconstructed plurality of MR images are transformed into k-space using an encoding operator comprising a compression operator.
4. A method as claimed in claim 1 , wherein the data consistency iteration step comprises minimising an Augmented Lagrangian with respect to the plurality of MR images.
5. A method as claimed in claim 1 , wherein the data consistency iteration step is of the form: {circumflex over (x)}=argmin X ƒ(x),
where ƒ(x) is a function defined as:
{data consistency component in k -space+(sparse consistency component in image space)}, and
μ is a regularization parameter.
6. A method as claimed in claim 1 , wherein the denoising iteration step comprises determining a difference between the reconstructed plurality of MR images and the sparse representations of the reconstructed plurality of MR images.
7. A method as claimed in claim 1 , wherein the denoising iteration step comprises minimising an Augmented Lagrangian with respect to sparse coefficients associated with a dictionary for the sparse representations of the plurality of MR images.
8. A method as claimed in claim 1 , wherein the denoising iteration step is of the form: {circumflex over (x)}=argmin X ƒ(x), where ƒ(x) is a function defined as:
{sparse consistency component in image
space
+
2
λ
μ
(sparse coefficients associated with a dictionary for the sparse representations of the MR images)} wherein λ and μ are regularization parameters.
9. A method as claimed in claim 1 , wherein using the similar patches across the plurality of MR images in a sparsifying operation comprises:
grouping the similar patches together to form a tensor;
performing a decomposition of the tensor so as to generate a sparse tensor; and
using the sparse tensor to obtain the sparse representations of the plurality of MR images.
10. A method as claimed in claim 9 , wherein the using the sparse tensor to obtain the sparse representations of the plurality of MR images comprises applying a threshold operator to the sparse tensor so as to extract the dominant components of the sparse tensor.
11. A method as claimed in claim 9 , wherein the grouping the similar patches together to form the tensor comprises:
unfolding the similar patches within each MR image to form a matrix having dimensions N×K, where N is the number of pixels or voxels in each patch, and K is the number of similar patches within each respective image; and
stacking the matrices obtained from the different reconstructed MR images together to form a tensor of dimensions N×K×L, where L is the number of reconstructed MR images.
12. A method as claimed in claim 9 , wherein the decomposition of the tensor so as to generate a sparse tensor comprises performing a singular value decomposition (SVD) on the tensor.
13. A method as claimed in claim 9 , wherein the k-space data is acquired using 3D MR acquisitions, and the 3D MR acquisitions are performed using a variable-density 3D Cartesian trajectory with spiral profile ordering.
14. A non-transitory computer readable medium having instructions recorded thereon, which, when executed by a computer, cause the computer to perform the method of claim 1 .
15. A magnetic resonance (MR) apparatus comprising:
a computing system;
a gradient system configured to apply a magnetic field gradient; and
an excitation system configured to apply an excitation pulse to a subject and to receive signals from the subject,
wherein the computing system is in communication with the excitation system and the gradient system for controlling these components, and is configured to receive the signals from the excitation system, and execute program code to control the gradient system and the excitation system to:
obtain a plurality of sets of k-space data each of the same image region of a subject, wherein the sets of k-space data were acquired with different contrasts; and
use a sparse image coding procedure to reconstruct a plurality of MR images each corresponding to one of the sets of k-space data by solving an optimization problem, wherein the optimization problem comprises:
a data consistency iteration step used to generate the reconstructed MR images, each of the reconstructed MR images corresponding to one of the sets of k-space data, and
a denoising iteration step applied to the reconstructed MR images generated during the data consistency iteration step, wherein the denoising iteration step comprises:
performing a 2D/3D block matching operation to identify similar patches across the reconstructed MR images; and
using the similar patches across the reconstructed MR images in a sparsifying operation to provide sparse representations of the reconstructed MR images, wherein the sparse representations are used as an input to the data consistency iteration step of the optimization problem.