Deep learning techniques for generating magnetic resonance images from spatial frequency data
Techniques for generating magnetic resonance (MR) images of a subject from MR data obtained by a magnetic resonance imaging (MRI) system, the techniques include: obtaining input MR spatial frequency data obtained by imaging the subject using the MRI system; generating an MR image of the subject from the input MR spatial frequency data using a neural network model comprising: a pre-reconstruction neural network configured to process the input MR spatial frequency data; a reconstruction neural network configured to generate at least one initial image of the subject from output of the pre-reconstruction neural network; and a post-reconstruction neural network configured to generate the MR image of the subject from the at least one initial image of the subject.
1. A method for generating magnetic resonance (MR) images of a subject from MR data obtained by a magnetic resonance imaging (MRI) system, the method comprising:
obtaining input MR spatial frequency data obtained by imaging the subject using the MRI system;
generating an MR image of the subject from the input MR spatial frequency data using a neural network model comprising:
a pre-reconstruction neural network configured to process the input MR spatial frequency data;
a reconstruction neural network configured to generate at least one initial image of the subject from processed MR spatial frequency data output by the pre-reconstruction neural network; and
a post-reconstruction neural network configured to generate the MR image of the subject from the at least one initial image of the subject output by the reconstruction neural network.
2. The method of claim 1 , wherein the input MR spatial frequency data is under-sampled relative to a Nyquist criterion.
3. The method of claim 1 , wherein points in the input MR spatial frequency data were obtained using a non-Cartesian sampling trajectory.
4. The method of claim 1 , wherein the pre-reconstruction neural network comprises a first neural network configured to suppress RF interference, the first neural network comprising one or more convolutional layers.
5. The method of claim 4 , wherein the pre-reconstruction neural network comprises a second neural network configured to suppress noise, the second neural network comprising one or more convolutional layers.
6. The method of claim 5 , wherein the pre-reconstruction neural network comprises a third neural network configured to perform line rejection, the third neural network comprising one or more convolutional layers.
7. The method of claim 1 , wherein the reconstruction neural network was trained to reconstruct MR images from spatial frequency MR data under-sampled relative to a Nyquist criterion.
8. The method of claim 1 , wherein the reconstruction neural network is configured to perform data consistency processing using a non-uniform Fourier transformation for transforming image data to spatial frequency data.
9. The method of claim 8 , wherein the reconstruction neural network is configured to perform data consistency processing using the non-uniform Fourier transformation at least in part by applying the non-uniform Fourier transformation on data by applying a de-apodization transformation, a fast Fourier transformation, and a gridding interpolation transformation to the data.
10. The method of claim 1 ,
wherein the MRI system comprises a plurality of RF coils;
wherein the at least one initial image of the subject comprises a plurality of images, each of the plurality of images generated from a portion of the input MR spatial frequency data collected by a respective RF coil in a plurality of RF coils;
wherein the post-reconstruction neural network comprises a first neural network configured to estimate a plurality of RF coil profiles corresponding to the plurality of RF coils,
the method further comprising:
generating the MR image of the subject using the plurality of MR images and the plurality of RF coil profiles.
11. The method of claim 1 ,
wherein the at least one initial image of the subject comprises a first set of one or more MR images and a second set of one or more MR images, and
wherein the post-reconstruction neural network comprises a second neural network for aligning the first set of MR images to the second set of MR images.
12. The method of claim 1 , wherein the post-reconstruction neural network comprises a neural network configured to suppress noise in the at least one initial image and/or at least one image obtained from the at least one initial image.
13. The method of claim 1 , wherein the pre-reconstruction neural network, the reconstruction neural network, and the post-reconstruction neural network are jointly trained with respect to a common loss function.
14. The method of claim 13 , where the common loss function is a weighted combination of a first loss function for the pre-reconstruction neural network, a second loss function for the reconstruction neural network, and a third loss function for the post-reconstruction neural network.
15. A magnetic resonance imaging (MRI) system, comprising:
a magnetics system having a plurality of magnetics components to produce magnetic fields for performing MRI; and
at least one processor configured to perform:
obtaining input MR spatial frequency data obtained by imaging the subject using the MRI system;
generating an MR image of the subject from the input MR spatial frequency data using a neural network model comprising:
a pre-reconstruction neural network configured to process the input MR spatial frequency data;
a reconstruction neural network configured to generate at least one initial image of the subject from processed MR spatial frequency data output by the pre-reconstruction neural network; and
a post-reconstruction neural network configured to generate the MR image of the subject from the at least one initial image of the subject output by the reconstruction neural network.
16. The MRI system of claim 15 , wherein the magnetics system comprises a permanent B 0 magnet configured to generate a B 0 magnetic field.
17. The MRI system of claim 16 , wherein the B 0 magnet comprises a plurality of concentric permanent magnet rings.
18. The MRI system of claim 15 , wherein the plurality of magnetics components include at least one permanent B 0 magnet configured to produce a B 0 field for an imaging region of the MRI system, the B 0 field having a strength between 50 milliTesla and 100 milliTesla.
19. The MRI system of claim 15 , wherein the plurality of magnetics components include at least one gradient coil.
20. At least one non-transitory computer readable storage medium storing processor-executable instructions that, when executed by at least one processor, cause the at least one processor to perform a method for generating magnetic resonance (MR) images of a subject from MR data obtained by a magnetic resonance imaging (MRI) system, the method comprising:
obtaining input MR spatial frequency data obtained by imaging the subject using the MRI system;
generating an MR image of the subject from the input MR spatial frequency data using a neural network model comprising:
a pre-reconstruction neural network configured to process the input MR spatial frequency data;
a reconstruction neural network configured to generate at least one initial image of the subject from processed MR spatial frequency data output by the pre-reconstruction neural network; and
a post-reconstruction neural network configured to generate the MR image of the subject from the at least one initial image of the subject output by the reconstruction neural network.