System and method for consistency-aware learnable multi-prior reconstruction for magnetic resonance imaging

Patent metadata

Field	Value
Publication number	US-12535547-B2
Application number	US-202418604294-A
Country	US
Kind code	B2
Filing date	Mar 13, 2024
Priority date	Apr 25, 2023
Publication date	Jan 27, 2026
Grant date	Jan 27, 2026

Abstract

Techniques for performing iterative MRI image reconstruction by learning complementary multi-prior knowledge from images, k-space data, and calibration data are disclosed. In one method, k-space data is obtained from an MRI scan. Image-space modifications are performed on the k-space data using a first neural network trained to operate on data in image space. The k-space data is converted from the frequency domain to a spatial domain to produce input image-space data. Using the first neural network, output image-space data is generated, which is then converted from the spatial domain to the frequency domain. K-space modifications are performed on the k-space data using a second neural network trained to operate on data in k-space. ACS are encoded using a third neural network to guide the second neural network in learning consistency-aware k-space correlations. The k-space data is converted from the frequency domain to the spatial domain to obtain a reconstructed image.

First claim

What is claimed is:

1 . A computer-implemented method of performing magnetic resonance imaging (MRI) image reconstruction, the computer-implemented method comprising:
obtaining k-space data from an MRI scan, wherein the k-space data is under-sampled in a frequency domain; for each of a set of iterations:
performing image-space modifications on the k-space data using a first neural network trained to operate on data in image space by:
converting the k-space data from the frequency domain to a spatial domain to produce input image-space data;
generating, using the first neural network, output image-space data by inputting the input image-space data into the first neural network; and
converting the output image-space data from the spatial domain to the frequency domain to obtain the k-space data incorporating the image-space modifications; and

performing k-space modifications on the k-space data using a second neural network trained to operate on data in k-space by inputting the k-space data into the second neural network to obtain the k-space data incorporating the k-space modifications; and
converting the k-space data from the frequency domain to the spatial domain to obtain a reconstructed image.

2 . The computer-implemented method of claim 1 , further comprising:
zero-filling the k-space data.

3 . The computer-implemented method of claim 1 , further comprising:
estimating a set of coil sensitivity maps based on the k-space data, wherein the image-space modifications are performed on the k-space data further using the set of coil sensitivity maps.

4 . The computer-implemented method of claim 3 , wherein the reconstructed image is obtained by using the set of coil sensitivity maps to compensate for an uneven signal reception strength for individual receiver coils used in a multi-coil setup.

5 . The computer-implemented method of claim 1 , further comprising:
for each of the set of iterations:
performing a frequency fusion operation on the k-space data by summing the k-space data incorporating the image-space modifications and the k-space data incorporating the k-space modifications.

6 . The computer-implemented method of claim 5 , wherein performing the frequency fusion operation includes using an under-sampling pattern to sum the k-space data incorporating the image-space modifications and the k-space data incorporating the k-space modifications.

7 . The computer-implemented method of claim 1 , wherein one or both of the first neural network and the second neural network were previously trained by:
obtaining training k-space data; performing the image-space modifications and the k-space modifications on the training k-space data and thereafter converting the training k-space data from the frequency domain to the spatial domain to obtain a training reconstructed image; comparing the training reconstructed image to a reference image to compute a reconstruction loss; and modifying weights associated with one or both of the first neural network and the second neural network based on the reconstruction loss.

8 . The computer-implemented method of claim 1 , wherein the k-space data incorporating the k-space modifications is obtained further using a surface data fidelity layer configured to reduce effects of data imperfections due to padding in the k-space domain.

9 . The computer-implemented method of claim 1 , wherein the k-space data incorporating the k-space modifications is obtained further using a calibration consistency module configured to encode calibration features from auto-calibration signals (ACS) and to guide the second neural network in learning consistency-aware k-space correlations, the calibration consistency module comprising a third neural network.

10 . A non-transitory computer-readable medium comprising instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising:
obtaining k-space data from a magnetic resonance imaging (MRI) scan, wherein the k-space data is under-sampled in a frequency domain; for each of a set of iterations:
performing image-space modifications on the k-space data using a first neural network trained to operate on data in image space by:
converting the k-space data from the frequency domain to a spatial domain to produce input image-space data;
generating, using the first neural network, output image-space data by inputting the input image-space data into the first neural network; and
converting the output image-space data from the spatial domain to the frequency domain to obtain the k-space data incorporating the image-space modifications; and

performing k-space modifications on the k-space data using a second neural network trained to operate on data in k-space by inputting the k-space data into the second neural network to obtain the k-space data incorporating the k-space modifications; and
converting the k-space data from the frequency domain to the spatial domain to obtain a reconstructed image.

11 . The non-transitory computer-readable medium of claim 10 , wherein the operations further comprise:
zero-filling the k-space data.

12 . The non-transitory computer-readable medium of claim 10 , wherein the operations further comprise:
estimating a set of coil sensitivity maps based on the k-space data, wherein the image-space modifications are performed on the k-space data further using the set of coil sensitivity maps.

13 . The non-transitory computer-readable medium of claim 12 , wherein the reconstructed image is obtained by using the set of coil sensitivity maps to compensate for an uneven signal reception strength for individual receiver coils used in a multi-coil setup.

14 . The non-transitory computer-readable medium of claim 10 , wherein the operations further comprise:
for each of the set of iterations:
performing a frequency fusion operation on the k-space data by summing the k-space data incorporating the image-space modifications and the k-space data incorporating the k-space modifications.

15 . The non-transitory computer-readable medium of claim 10 , wherein the k-space data incorporating the k-space modifications is obtained further using a calibration consistency module configured to encode calibration features from auto-calibration signals (ACS) and to guide the second neural network in learning consistency-aware k-space correlations, the calibration consistency module comprising a third neural network.

16 . The non-transitory computer-readable medium of claim 10 , wherein the operations further comprise:
training one or both of the first neural network and the second neural network by:
obtaining training k-space data;
performing the image-space modifications and the k-space modifications on the training k-space data and thereafter converting the training k-space data from the frequency domain to the spatial domain to obtain a training reconstructed image;
comparing the training reconstructed image to a reference image to compute a reconstruction loss; and
modifying weights associated with one or both of the first neural network and the second neural network based on the reconstruction loss.

17 . A system comprising:
one or more processors; and a computer-readable medium comprising instructions that, when executed by the one or more processors, cause the one or more processors to perform operations comprising:
obtaining k-space data from a magnetic resonance imaging (MRI) scan, wherein the k-space data is under-sampled in a frequency domain;
for each of a set of iteration

Assignees

• Univ Hong Kong Chinese

Inventors

• Zhang Liping
• Chen Weitian

Classifications

• G06T2211/441

  AI-based methods, deep learning or artificial neural networks · CPC title

• G06T2211/424

  Iterative · CPC title

• G06T11/006

  Physics · mapped topic

• G06T11/005

  Physics · mapped topic

• G01R33/58

  Calibration of imaging systems, e.g. using test probes {, Phantoms; Calibration objects or fiducial markers such as active or passive RF coils surrounding an MR active material} · CPC title