Methods, systems, and computer readable media for using a trained adversarial network for performing retrospective magnetic resonance imaging (MRI) artifact correction

Patent metadata

Field	Value
Publication number	US-11360180-B2
Application number	US-202017127366-A
Country	US
Kind code	B2
Filing date	Dec 18, 2020
Priority date	Dec 19, 2019
Publication date	Jun 14, 2022
Grant date	Jun 14, 2022

Abstract

A method for performing retrospective magnetic resonance imaging (MRI) artifact correction includes receiving, as input, an MRI image having at least one artifact; using a trained adversarial network for performing retrospective artifact correction on the MRI image, wherein the trained adversarial network is trained using unpaired artifact-free MRI images and artifact-containing MRI images; and outputting, by the trained adversarial network, a derivative MRI image related to the input, wherein the at least one artifact is corrected in the derivative MRI image.

First claim

What is claimed is:

1. A method for performing retrospective magnetic resonance imaging (MRI) artifact correction, the method comprising:
receiving, as input, a magnetic resonance imaging (MRI) image having at least one artifact;
using a trained adversarial network for performing retrospective artifact correction on the MRI image, wherein the trained adversarial network is trained using unpaired artifact-free MRI images and artifact-containing MRI images, wherein the trained adversarial network is trained using a first autoencoder for translating images of a first image domain to a second image domain and a second autoencoder for translating images of the second image domain to the first image domain; and
outputting, by the trained adversarial network, a derivative MRI image related to the input, wherein the at least one artifact is corrected in the derivative MRI image.

2. The method of claim 1 wherein the first autoencoder includes a first artifact encoder for encoding artifact information associated with a first image of the first image domain and a first content encoder for encoding content information associated with a first image of the first image domain and wherein the second autoencoder includes a second artifact encoder for encoding artifact information associated with a first image of the second image domain and a second content encoder for encoding content information associated with a first image of the second image domain.

3. The method of claim 1 wherein the first autoencoder includes a decoder for decoding encoded content information and encoded artifact information into a second image of the first image domain or a second image of the second image domain.

4. The method of claim 3 wherein the encoded content information used by the decoder is associated with the first image of the first image domain and wherein the encoded artifact information used by the decoder is associated with the first image of the second image domain.

5. The method of claim 1 wherein the first image domain represents a group of artifact-containing MRI images and the second image domain represents a group of artifact-free MRI images.

6. The method of claim 1 wherein the trained adversarial network is trained using a first discriminator for distinguishing a real image of the first image domain from an adversarial network generated image of the first image domain and a second discriminator for distinguishing a real image of the second image domain from an adversarial network generated image of the second image domain.

7. The method of claim 1 wherein the trained adversarial network is trained using a pixel-wise consistency loss function for enforcing identity translation mapping, two least-squares loss functions for adversarial learning, or a multi-scale content consistency loss function based on pixel, low-level content features, and high-level content features between images.

8. The method of claim 1 wherein the MRI image having at least one artifact is a T1-weighted MRI image or a T2-weighted MRI image.

9. A system for performing retrospective magnetic resonance imaging (MRI) artifact correction, the system comprising:
at least one computing platform including at least one processor; and
a trained adversarial network executable by the at least one processor for:
receiving, as input, a magnetic resonance imaging (MRI) image having at least one artifact;
using a trained adversarial network for performing retrospective artifact correction on the MRI image, wherein the trained adversarial network is trained using unpaired artifact-free MRI images and artifact-containing MRI images, wherein the trained adversarial network is trained using a first autoencoder for translating images of a first image domain to a second image domain and a second autoencoder for translating images of the second image domain to the first image domain; and
outputting, by the trained adversarial network, a derivative MRI image related to the input, wherein the at least one artifact is corrected in the derivative MRI image.

10. The system of claim 9 wherein the first autoencoder includes a first artifact encoder for encoding artifact information associated with a first image of the first image domain and a first content encoder for encoding content information associated with a first image of the first image domain and wherein the second autoencoder includes a second artifact encoder for encoding artifact information associated with a first image of the second image domain and a second content encoder for encoding content information associated with a first image of the second image domain.

11. The system of claim 9 wherein the first autoencoder includes a decoder for decoding encoded content information and encoded artifact information into a second image of the first image domain or a second image of the second image domain.

12. The system of claim 11 wherein the encoded content information used by the decoder is associated with the first image of the first image domain and wherein the encoded artifact information used by the decoder is associated with the first image of the second image domain.

13. The system of claim 9 wherein the first image domain represents a group of artifact-containing MRI images and the second image domain represents a group of artifact-free MRI images.

14. The system of claim 9 wherein the trained adversarial network is trained using a first discriminator for distinguishing a real image of the first image domain from an adversarial network generated image of the first image domain and a second discriminator for distinguishing a real image of the second image domain from an adversarial network generated image of the second image domain.

15. The system of claim 9 wherein the trained adversarial network is trained using a pixel-wise consistency loss function for enforcing identity translation mapping, two least-squares loss functions for adversarial learning, or a multi-scale content consistency loss function based on pixel, low-level content features, and high-level content features between images.

16. The system of claim 9 wherein the MRI image having at least one artifact is a T1-weighted MRI image or a T2-weighted MRI image.

17. A non-transitory computer readable medium having stored thereon executable instructions that when executed by at least one processor of at least one computer cause the at least one computer to perform steps comprising:
receiving, as input, a magnetic resonance imaging (MRI) image having at least one artifact;
using a trained adversarial network for performing retrospective artifact correction on the MRI image, wherein the trained adversarial network is trained using unpaired artifact-free MRI images and artifact-containing MRI images, wherein the trained adversarial network is trained using a first autoencoder for translating images of a first image domain to a second image domain and a second autoencoder for translating images of the second image domain to the first image domain; and
outputting, by the trained adversarial network, a derivative MRI image related to the input, wherein the at least one artifact is corrected in the derivative MRI image.

Assignees

• Univ North Carolina Chapel Hill

Inventors

• Yap Pew-Thian
• Liu Siyuan

Classifications

• A61B5/055

  involving electronic [EMR] or nuclear [NMR] magnetic resonance, e.g. magnetic resonance imaging · CPC title

• G06N3/047

  Probabilistic or stochastic networks · CPC title

• G06N3/048

  Activation functions · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06N3/0475

  Generative networks · CPC title