Self ensembling 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 including: obtaining input MR data obtained by imaging the subject using the MRI system; generating a plurality of transformed input MR data instances by applying a respective first plurality of transformations to the input MR data; generating a plurality of MR images from the plurality of transformed input MR data instances and the input MR data using a non-linear MR image reconstruction technique; generating an ensembled MR image from the plurality of MR images at least in part by: applying a second plurality of transformations to the plurality of MR images to obtain a plurality of transformed MR images; and combining the plurality of transformed MR images to obtain the ensembled MR image; and outputting the ensembled MR image.
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 data obtained by imaging the subject using the MRI system;
generating a plurality of transformed input MR data instances by applying a respective first plurality of transformations to the input MR data;
generating a plurality of MR images from the plurality of transformed input MR data instances and the input MR data using a non-linear MR image reconstruction technique;
generating an ensembled MR image from the plurality of MR images at least in part by:
applying a second plurality of transformations to the plurality of MR images to obtain a plurality of transformed MR images, wherein the second plurality of transformations are selected to reverse effects of applying the first plurality of transformations; and
combining the plurality of transformed MR images to obtain the ensembled MR image; and
outputting the ensembled MR image.
2. The method of claim 1 , wherein using the non-linear MR image reconstruction technique comprises applying a neural network model to the transformed input MR data instances to obtain the plurality of MR images.
3. The method of claim 1 , wherein using the non-linear MR image reconstruction technique comprises using a compressed sensing technique.
4. The method of claim 1 , wherein applying a first plurality of transformations to the input MR data comprises applying one or more of a constant phase shift transformation, a linear phase shift transformation, a complex conjugation transformation, a rotation transformation, a transpose transformation, and/or a reflection transformation.
5. The method of claim 1 , wherein:
applying the first plurality of transformations to the input MR data comprises applying the first plurality of transformations to the input MR data in a spatial frequency domain;
applying the second plurality of transformations to the plurality of MR images comprises applying the second plurality of transformations to the plurality of MR images in an image domain; and
the second plurality of transformations are selected to reverse effects of applying the first plurality of transformations in the spatial frequency domain.
6. The method of claim 1 , wherein the input MR data comprises first spatial frequency MR data (y i ) for generating an image for a first anatomy slice of a subject and second spatial frequency MR data (y i+1 ) for generating an image for a second anatomy slice of the subject, and wherein:
generating the plurality of transformed input MR data instances comprises generating a first transformed input MR data instance (y i +1 ) by adding the second spatial frequency MR data to the first spatial frequency MR data;
generating the plurality of MR images comprises:
generating a first MR image (x i +1 ) from the first transformed data instance (y i +1 ), and
generating a second MR image (x i+1 ) from the second MR spatial frequency data (y i+1 ); and
generating the ensembled MR image comprises:
subtracting the second MR image from the first MR image (x i +1 −x i+1 ).
7. The method of claim 1 , wherein the input MR data comprises first spatial frequency MR data for generating an image for a first anatomy slice of a subject and second spatial frequency MR data for generating one or more images for one or more other anatomy slices of the subject, and wherein:
generating the plurality of transformed input MR data instances comprises generating a first transformed input MR data instance by combining the first spatial frequency MR data and the second spatial frequency MR data;
generating the plurality of MR images comprises:
generating a first MR image from the first transformed input MR data instance, and
generating one or more second MR images from the second spatial frequency MR data; and
generating the ensembled MR image comprises:
subtracting the one or more second MR images from the first MR image.
8. The method of claim 1 , wherein the first plurality of transformations includes a complex conjugation transformation and the second plurality of transformations includes a reflection transformation.
9. The method of claim 1 , wherein combining the plurality of transformed MR images to obtain the ensembled MR image comprises computing the ensembled MR image as a weighted average of the plurality of transformed MR images.
10. 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 data obtained by imaging the subject using the MRI system;
generating a plurality of transformed input MR data instances by applying a respective first plurality of transformations to the input MR data;
generating a plurality of MR images from the plurality of transformed input MR data instances and the input MR data using a non-linear MR image reconstruction technique;
generating an ensembled MR image from the plurality of MR images at least in part by:
applying a second plurality of transformations to the plurality of MR images to obtain a plurality of transformed MR images, wherein the second plurality of transformations are selected to reverse effects of applying the first plurality of transformations; and
combining the plurality of transformed MR images to obtain the ensembled MR image; and
outputting the ensembled MR image.
11. The at least one non-transitory computer readable storage medium of claim 10 , wherein using the non-linear MR image reconstruction technique comprises applying a neural network model to the transformed input MR data instances to obtain the plurality of MR images.
12. The at least one non-transitory computer readable storage medium of claim 10 , wherein using the non-linear MR image reconstruction technique comprises using a compressed sensing technique.
13. The at least one non-transitory computer readable storage medium of claim 10 , wherein applying a first plurality of transformations to the input MR data comprises applying one or more of a constant phase shift transformation, a linear phase shift transformation, a complex conjugation transformation, a rotation transformation, a transpose transformation, and/or a reflection transformation.
14. The at least one non-transitory computer readable storage medium of claim 10 , wherein:
applying the first plurality of transformations to the input MR data comprises applying the first plurality of transformations to the input MR data in a spatial frequency domain;
applying the second plurality of transformations to the plurality of MR images comprises applying the second plurality of transformations to the plurality of MR images in an image domain; and
the second plurality of transformations are selected to reverse effects of applying the first plurality of transformations in the spatial frequency domain.
15. The at least one non-transitory computer readable storage medium of claim 10 , wherein the input MR data comprises first spatial frequency MR data (y i ) for generating an image for a first anatomy slice of a subject and second spatial frequency MR data (y i+1 ) for generating an image for a second anatomy slice of the subject, and wherein:
generating the plurality of transformed input MR data instances comprises generating a first transformed input MR data instance (y i +1 ) by adding the second spatial frequency MR data to the first spatial frequency MR data;
generating the plurality of MR images comprises:
generating a first MR image (x i +1 ) from the first transformed data instance (y i +1 ), and
generating a second MR image (x i+1 ) from the second MR spatial frequency data (y i+1 ); and
generating the ensembled MR image comprises:
subtracting the second MR image from the first MR image (x i +1 −x i+1 ).
16. The at least one non-transitory computer readable storage medium of claim 10 , wherein the input MR data comprises first spatial frequency MR data for generating an image for a first anatomy slice of a subject and second spatial frequency MR data for generating one or more images for one or more other anatomy slices of the subject, and wherein:
generating the plurality of transformed input MR data instances comprises generating a first transformed input MR data instance by combining the first spatial frequency MR data and the second spatial frequency MR data;
generating the plurality of MR images comprises:
generating a first MR image from the first transformed input MR data instance, and
generating one or more second MR images from the second spatial frequency MR data; and
generating the ensembled MR image comprises:
subtracting the one or more second MR images from the first MR image.
17. The at least one non-transitory computer readable storage medium of claim 10 , wherein the first plurality of transformations includes a complex conjugation transformation and the second plurality of transformations includes a reflection transformation.
18. The at least one non-transitory computer readable storage medium of claim 10 , wherein combining the plurality of transformed MR images to obtain the ensembled MR image comprises computing the ensembled MR image as a weighted average of the plurality of transformed MR images.
19. A magnetic resonance imaging (MRI) system configured to capture a magnetic resonance (MR) image, the 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 data obtained by imaging a subject using the MRI system;
generating a plurality of transformed input MR data instances by applying a respective first plurality of transformations to the input MR data;
generating a plurality of MR images from the plurality of transformed input MR data instances and the input MR data using a non-linear MR image reconstruction technique;
generating an ensembled MR image from the plurality of MR images at least in part by:
applying a second plurality of transformations to the plurality of MR images to obtain a plurality of transformed MR images, wherein the second plurality of transformations are selected to reverse effects of applying the first plurality of transformations; and
combining the plurality of transformed MR images to obtain the ensembled MR image; and
outputting the ensembled MR image.
20. The MRI system of claim 19 , wherein the plurality of magnetics components comprises at least one permanent B 0 magnet configured to generate a B 0 magnetic field, the at least one permanent B 0 magnet comprising a plurality of permanent magnet rings.