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, from magnitude MR images, training data for training a neural network model to process data to be collected by a target MRI system, the training data comprising the magnitude MR images and synthetic spatial frequency data generated from the magnitude MR images in part by using characteristics of the target MRI system, the method comprising:
using at least one computer hardware processor to perform:
(A) obtaining a reference MR volume comprising one or more magnitude MR images;
(B) updating the reference MR volume, using a target field of view and/or target image resolution for the target MRI system, to obtain an updated MR volume;
(C) generating synthetic phase and adding the generated synthetic phase to the updated MR volume to obtain a target MR volume; and
(D) generating, from the target MR volume and using one or more of the characteristics of the target MRI system, multiple sets of spatial frequency data to be used as part of the training data for training the neural network model to process the data to be collected by the target MRI system.
2. The method of claim 1 , wherein the one or more magnitude MR images include one or more magnitude MR images obtained by a clinical MRI system different from the target MRI system.
3. The method of claim 1 , wherein the clinical MRI system is a high-field MRI system having a B0 field strength at or above 0.5 T and the target MRI system is a low-field MRI system having a B0 field strength between 0.02 T and 0.2 T.
4. The method of claim 1 , wherein (B) further comprises cropping and resampling the reference MR volume using the target field of view and the target image resolution to obtain the updated MR volume.
5. The method of claim 1 , further comprising:
after performing (B) and prior to performing (C), applying an affine transformation to the updated MR volume to simulate a possible position and/or orientation of a patient's anatomy within the target MRI system.
6. The method of claim 1 , further comprising:
after performing (B) and prior to performing (C), applying a deformation to the updated MR volume to simulate effect of inhomogeneity of the target MRI system's B0 field, eddy currents, and/or encoding error of the target MRI system.
7. The method of claim 1 , further comprising:
after performing (B) and prior to performing (C), applying a histogram augmentation the updated MR volume.
8. The method of claim 1 , wherein generating the synthetic phase comprises:
generating the synthetic phase using a linear combination of spherical harmonic basis functions in which the spherical harmonic basis functions are weighted by coefficients sampled at random.
9. The method of claim 1 , wherein the characteristics of the target MRI system comprise one or more of: size of the field of view of the target MRI system, sampling patterns to be used by the target MRI system during imaging, number of RF coils in the target MRI system configured to detect MR data, geometry and sensitivity of the RF coils in the target MRI system, pulse correlation among signals received by the RF coils of the target MRI system, external RF interference that the target MRI system is expected to experience during operation, internal RF interference that the target MRI system is expected to experience during operation, pulse sequences to be used by the target MRI during imaging, and field strength of the target MRI system.
10. The method of claim 1 , wherein (D) further comprises:
generating RF artefacts to simulate one or more types of RF artefacts expected to be observed during application of a particular pulse sequence by the target MRI system; and
applying the generated RF artefacts to the target MR volume.
11. The method of claim 1 , wherein (D) further comprises:
generating an affine transformation to simulate effect of patient motion during application of a particular pulse sequence by the target MRI system; and
applying the affine transformation to the target MR volume.
12. The method of claim 1 ,
wherein the target MRI system includes multiple RF coils, and
wherein (D) further comprises:
generating an RF coil sensitivity profile for each of the multiple RF coils; and
applying the generated RF coil sensitivity profiles to the target MR volume to obtain multiple MR volumes.
13. The method of claim 12 , further comprising:
determining a coil correlation matrix to model effect coupling and/or inductance among the multiple RF coils; and
applying the coil correlation matrix to the multiple MR volumes.
14. The method of claim 13 , further comprising:
generating correlated Gaussian noise using the coil correlation matrix; and
applying the correlated Gaussian noise to the multiple MR volumes to obtain noise-corrupted coil-weighted MR volumes.
15. The method of claim 14 , further comprising:
applying a non-uniform Fourier transformation to the noise-corrupted coil-weighted MR volumes to get spatial frequency data to be used as part of the training data.
16. The method of claim 1 , further comprising:
repeating acts (C), and (D) to obtain further synthetic spatial frequency data from the reference MR volume.
17. The method of claim 1 , further comprising using the generated training data to train train the neural network model to process the data to be collected by the target MRI system.
18. The method of claim 17 , wherein training the neural network model comprises training the neural network model to reconstruct MR images from spatial frequency data.
19. A system for generating, from magnitude MR images, training data for training a neural network model to process data to be collected by a target MRI system, the training data comprising the magnitude MR images and synthetic spatial frequency data generated from the magnitude MR images in part by using characteristics of the target MRI system, the system comprising:
at least one computer hardware processor configured to perform:
(A) obtaining a reference MR volume comprising one or more magnitude MR images;
(B) updating the reference MR volume, using a target field of view and/or target image resolution for the target MRI system, to obtain an updated MR volume;
(C) generating synthetic phase and adding the generated synthetic phase to the updated MR volume to obtain a target MR volume; and
(D) generating, from the target MR volume and using one or more of the characteristics of the target MRI system, multiple sets of spatial frequency data to be used as part of the training data for training the neural network model to process the data to be collected by the target MRI system.
20. At least one non-transitory computer-readable storage medium storing processor-executable instructions that, when executed by at least one computer hardware processor, cause the at least one computer hardware processor to perform a method for generating, from magnitude MR images, training data for training a neural network model to process data to be collected by a target MRI system, the training data comprising the magnitude MR images and synthetic spatial frequency data generated from the magnitude MR images in part by using characteristics of the target MRI system, the method comprising:
(A) obtaining a reference MR volume comprising one or more magnitude MR images;
(B) updating the reference MR volume, using a target field of view and/or target image resolution for the target MRI system, to obtain an updated MR volume;
(C) generating synthetic phase and adding the generated synthetic phase to the updated MR volume to obtain a target MR volume; and
(D) generating, from the target MR volume and using one or more of the characteristics of the target MRI system, multiple sets of spatial frequency data to be used as part of the training data for training the neural network model to process the data to be collected by the target MRI system.