Methods and systems for real-time 3D MRI

Patent metadata

Field	Value
Publication number	US-11294014-B2
Application number	US-202016813727-A
Country	US
Kind code	B2
Filing date	Mar 9, 2020
Priority date	Mar 7, 2019
Publication date	Apr 5, 2022
Grant date	Apr 5, 2022

Abstract

Among the various aspects of the present disclosure is the provision of methods and systems for real-time 3D MRI that combines dynamic keyhole data sharing with super-resolution imaging methods to improve real-time 3D MR images in the presence of motion.

First claim

What is claimed is:

1. A computer-aided method of generating a real-time 3D magnetic resonance (MR) image of a subject comprising:
obtaining a keyhole 3D MR image of the subject, the keyhole 3D MR image comprising a k-space central dataset;
transforming, using a computing device, the keyhole 3D MR image into a super-resolution 3D MR image using a deep-learning super-resolution generative (SRG) model;
extracting, using the computing device, a respiratory phase of the super-resolution 3D MR image;
combining, using the computing device, an SR k-space central dataset corresponding to the super-resolution 3D MR image with a respiratory phase-matched k-space peripheral dataset retrieved from a stored library of k-space peripheral datasets to produce a combined k-space dataset; and
reconstructing, using the computing device, the real-time 3D MR image from the combined k-space dataset.

2. The method of claim 1 , wherein the k-space central dataset comprises reduced k-space data relative to a high-resolution 3D MR image, the reduced k-space data comprising reduced kx data, reduced ky data, reduced kz data, and any combination thereof.

3. The method of claim 2 , wherein each peripheral k-space dataset of the library comprises a k-space dataset corresponding to a high-resolution 3D MR image with the k-space central dataset removed.

4. The method of claim 3 , further comprising obtaining the stored library of k-space peripheral datasets by:
obtaining a plurality of high-resolution 3D MR images of the subject;
extracting, using the computing device, an index respiratory phase from each high-resolution 3D MR image;
removing, using the computing device, each k-space central dataset from each k-state dataset of each of high-resolution 3D MR images to produce each k-space peripheral dataset;
assembling, using the computing device, each k-space peripheral dataset and each corresponding index respiratory phase from each high-resolution 3D MR image to produce the stored library of k-space peripheral datasets.

5. The method of claim 1 , wherein the deep-learning super-resolution generative (SRG) model comprises a generator portion of a cascading deep learning model trained using a generative adversarial network framework with a training dataset of matched low resolution and high resolution MR images.

6. The method of claim 1 , further comprising denoising, using the computing device, the keyhole 3D MR image using a denoising autoencoder (DAE), the DAE comprising a convolutional neural network (CNN).

7. The method of claim 1 , wherein extracting the respiratory phase further comprises extracting, using the computing device, a profile of the diaphragm on a middle coronal slice of the super-resolution 3D MR image.

8. The method of claim 1 , wherein extracting the respiratory phase further comprises extracting, using the computing device, a 1D Fourier transform of a kx line at a ky-kz center of the k-space central dataset.

9. The method of claim 1 , wherein the real-time 3D MR image is reconstructed using a 3D Fourier-transform of the combined k-space dataset.

10. The method of claim 1 , wherein the real-time 3D MR image of the subject includes a target region of the subject, the target region comprising a tumor, and organ, a tissue, any portion thereof, and any combination thereof.

11. The method of claim 1 , wherein the image is generated in the presence of motion.

12. A computer-implemented method of treating a subject in need thereof comprising:
generating real-time 3D MR images of the subject by:
obtaining a keyhole 3D MR image of the subject, the keyhole 3D MR image comprising a k-space central dataset;
transforming, using a computing device, the keyhole 3D MR image into a super-resolution 3D MR image using a deep-learning super-resolution generative (SRG) model;
extracting, using the computing device, a respiratory phase of the super-resolution 3D MR image;
combining, using the computing device, an SR k-space central dataset corresponding to the super-resolution 3D MR image with a respiratory phase-matched k-space peripheral dataset retrieved from a stored library of k-space peripheral datasets to produce a combined k-space dataset; and
reconstructing, using the computing device, the real-time 3D MR image from the combined k-space dataset; and

at least one of: managing motion of a subject or a tumor in the subject using the generated real-time 3D MR images, gating a treatment of a subject or a tumor in the subject using the real-time 3D MR images, and any combination thereof.

13. The method of claim 12 , wherein the k-space central dataset comprises reduced k-space data relative to a high-resolution 3D MR image, the reduced k-space data comprising reduced kx data, reduced ky data, reduced kz data, and any combination thereof.

14. The method of claim 13 , wherein each peripheral k-space dataset of the library comprises a k-space dataset corresponding to a high-resolution 3D MR image with the k-space central dataset removed.

15. The method of claim 14 , further comprising obtaining the stored library of k-space peripheral datasets by:
obtaining a plurality of high-resolution 3D MR images of the subject;
extracting, using the computing device, an index respiratory phase from each high-resolution 3D MR image;
removing, using the computing device, each k-space central dataset from each k-state dataset of each of high-resolution 3D MR images to produce each k-space peripheral dataset;
assembling, using the computing device, each k-space peripheral dataset and each corresponding index respiratory phase from each high-resolution 3D MR image to produce the stored library of k-space peripheral datasets.

16. A system for generating a real-time 3D MR image of a subject, the system comprising:
a computing device comprising at least one processor, the at least one processor configured to:
obtain a keyhole 3D MR image of the subject, the keyhole 3D MR image comprising a k-space central dataset;
transform, using a computing device, the keyhole 3D MR image into a super-resolution 3D MR image using a deep-learning super-resolution generative (SRG) model;
extract, using the computing device, a respiratory phase of the super-resolution 3D MR image;
combine, using the computing device, an SR k-space central dataset corresponding to the super-resolution 3D MR image with a respiratory phase-matched k-space peripheral dataset retrieved from a stored library of k-space peripheral datasets to produce a combined k-space dataset; and
reconstruct, using the computing device, the real-time 3D MR image from the combined k-space dataset.

17. The system of claim 16 , wherein the k-space central dataset comprises reduced k-space data relative to a high-resolution 3D MR image, the reduced k-space data comprising reduced kx data, reduced ky data, reduced kz data, and any combination thereof.

18. The system of system 17 , wherein each peripheral k-space dataset of the library comprises a k-space dataset corresponding to a high-resolution 3D MR image with the k-space central dataset removed.

19. The system of claim 18 , further comprising a magnetic resonance imaging (MRI) scanner obtaining the keyhole 3D MR image of the subject.

Assignees

• Kim Taeho
• Park Chunjoo
• Univ Washington

Inventors

• Kim Taeho
• Park Chunjoo

Classifications

• G01R33/56509 Primary

  due to motion, displacement or flow, e.g. gradient moment nulling (G01R33/567 takes precedence) · CPC title

• A61B5/055 Primary

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

• G06N3/047

  Probabilistic or stochastic networks · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06N3/09

  Supervised learning · CPC title