Magnetic resonance imaging reconstruction using machine learning and motion compensation
The disclosure relates to MRI reconstruction of multiple MRI measurement datasets acquired throughout a measurement time duration. Patient motion can occur during the measurement time duration. Warping operators, sometimes also referred to as motion field, are incorporated into an iterative optimization of the MRI reconstruction.
1. A computer-implemented method of reconstructing a sequence of magnetic resonance imaging (MRI) images, comprising:
obtaining, via one or more processors, a sequence of MRI measurement datasets that have been acquired during a measurement time duration during which patient movement occurs, each respective one of MRI measurement datasets of the sequence of MRI measurement datasets being acquired using a respective under-sampling trajectory in k-space and a receiver coil array; and
performing, via the one or more processors, an iterative process to obtain a sequence of reconstructed MRI images based on the MRI measurement datasets,
wherein the iterative process comprises, for each iteration of multiple iterations of the iterative process, a regularization operation and a data-consistency operation to obtain a respective current MRI image,
wherein the data-consistency operation is based on differences between the MRI measurement datasets and synthesized MRI measurement datasets, the synthesized MRI measurement datasets being based on a k-space representation of a prior MRI image of the multiple iterations, an undersampling trajectory, and a sensitivity map associated with the receiver coil array, and
wherein an input to the regularization operation comprises, for each iteration of the multiple iterations, a concatenation of multiple prior images obtained from a previous iteration of the multiple iterations, the multiple prior images being associated with multiple motion states of the patient movement and with multiple points in time throughout the measurement time duration, the multiple prior images being warped from their respective motion states to a reference motion state.
2. The computer-implemented method of claim 1 , wherein the multiple prior images are warped to the reference motion state by applying warping operators between the multiple motion states and the reference motion state, the method further comprising:
performing, via the one or more processors, a further image reconstruction of a sequence of preliminary MRI images based on the MRI measurement datasets, and
determining, via the one or more processors, the warping operators from a sequence of motion states of the movement of the patient to and from the reference motion state based on the sequence of preliminary MRI images.
3. The computer-implemented method of claim 2 , wherein the further image reconstruction is based on representations of the MRI measurement datasets having a reduction in resolution.
4. The computer-implemented method of claim 2 , wherein the warping operators are determined based on image registrations between pairs of preliminary MRI images of the sequence of preliminary MRI images.
5. The computer-implemented method of claim 2 , further comprising:
for at least one iteration of the multiple iterations of the iterative process, adjusting the warping operators based on at least one of the prior image and the current image of the respective iteration.
6. The computer-implemented method of claim 2 , wherein:
the warping operators are determined using a neural network,
the regularization operation is implemented by a further neural network, and
the neural network and the further neural network are trained end-to-end.
7. The computer-implemented method of claim 6 , wherein:
the warping operators are determined using a neural network and based on image registrations between pairs of preliminary MRI images of the sequence of preliminary MRI images,
the regularization operation and the image registrations are implemented by a further neural network, and
the neural network and the further neural network are trained end-to-end.
8. The computer-implemented method of claim 1 , wherein an output of the regularization operation comprises a concatenation of multiple current MRI images, the method further comprising:
un-warping each one of multiple current MRI images from the reference motion state to a respective motion state associated with the patient movement.
9. The computer-implemented method of claim 1 , wherein the regularization operation is implemented by a trained neural network, and
wherein the concatenation of the multiple prior images corresponds to a stacking of the multiple prior images along a channel dimension of the trained neural network.
10. The computer-implemented method of claim 9 , wherein the regularization operation is performed as a joint regularization to determine, via the neural network, interdependencies between the concatenation of the multiple prior images.
11. The computer-implemented method of claim 1 , wherein the regularization operation comprises one or more operations that are impacted by pixel or voxel values of two or more images from among the concatenation of the multiple prior images.
12. The computer-implemented method of claim 1 , wherein the regularization operation discriminates between MRI images and aliasing artifacts due to undersampling.
13. A computer-implemented method of reconstructing a sequence of magnetic resonance imaging (MRI) images, the method comprising:
obtaining, via one or more processors, a sequence of MRI measurement datasets that have been acquired during a measurement time duration during which a patient movement occurs, each respective one of MRI measurement datasets of the sequence of MRI measurement datasets being acquired using an under-sampling trajectory in k-space and a receiver coil array; and
performing, via one or more processors, an iterative process to obtain a sequence of reconstructed MRI images based on the MRI measurement datasets,
wherein the iterative process comprises, for each iteration of multiple iterations of the iterative process, a regularization operation and a data-consistency operation to obtain a respective current MRI image,
wherein the data-consistency operation is based on differences between the MRI measurement datasets and synthesized MRI measurement datasets, the synthesized MRI measurement datasets being based on a k-space representation of a prior image of the multiple iterations, an undersampling trajectory, and a sensitivity map associated with the receiver coil array,
wherein an input to the regularization operation comprises, for each iteration of the multiple iterations, a concatenation of multiple prior images obtained from a previous iteration of the multiple iterations, the multiple prior images being associated with multiple motion states of the patient movement and with multiple points in time throughout the measurement time duration, and
wherein the regularization operation comprises, for each iteration of the multiple iterations, a first convolution of the input in spatial domain, a second convolution of the input in time domain, and a third convolution of the input in motion-state domain.
14. The computer-implemented method of claim 13 ,
wherein the first convolution is a two-dimensional (2-D) or a three-dimensional (3-D) convolution, and
wherein the second convolution and the third convolution are one-dimensional (1-D) convolutions, respectively.
15. The computer-implemented method of claim 13 , further comprising:
separately training components of the regularization operation associated with the first convolution, the second convolution, and the third convolution.
16. The computer-implemented method of claim 15 ,
wherein a resolution in at least one of the time domain and the motion-state domain is reduced when training the component of the regularization operation associated with the first convolution.
17. The computer-implemented method of claim 13 , further comprising:
increasing a spatial resolution of the current MRI images over the multiple iterations.
18. The computer-implemented method of claim 13 , further comprising:
increasing a temporal resolution of the current MRI images over the multiple iterations.
19. The computer-implemented method of claim 13 , further comprising:
increasing a motion-state resolution of the current MRI images over the multiple iterations.
20. A computer-readable storage medium comprising program code that, when executed by at least one processor, cause the at least one processor to reconstruct a sequence of magnetic resonance imaging (MRI) images by:
obtaining a sequence of MRI measurement datasets that have been acquired during a measurement time duration during which patient movement occurs, MRI measurement datasets of the sequence of MRI measurement datasets being acquired using a respective under-sampling trajectory in k-space and a receiver coil array; and
performing an iterative process to obtain a sequence of reconstructed MRI images based on the MRI measurement datasets,
wherein the iterative process comprises, for each iteration of multiple iterations of the iterative process, a regularization operation and a data-consistency operation to obtain a respective current MRI image,
wherein the data-consistency operation is based on differences between the MRI measurement datasets and synthesized MRI measurement datasets, the synthesized MRI measurement datasets being based on a k-space representation of a prior MRI image of the multiple iterations, an undersampling trajectory, and a sensitivity map associated with the receiver coil array, and
wherein an input to the regularization operation comprises, for each iteration of the multiple iterations, a concatenation of multiple prior images obtained from a previous iteration of the multiple iterations, the multiple prior images being associated with multiple motion states of the patient movement and with multiple points in time throughout the measurement time duration, the multiple prior images being warped from their respective motion states to a reference motion state.