Methods and systems for removing spike noise in magnetic resonance imaging

Patent metadata

Field	Value
Publication number	US-10705170-B1
Application number	US-201916277922-A
Country	US
Kind code	B1
Filing date	Feb 15, 2019
Priority date	Feb 15, 2019
Publication date	Jul 7, 2020
Grant date	Jul 7, 2020

Abstract

Spike noise in a k-space dataset acquired in magnetic resonance imaging may be removed by generating a mask including a set of data points which constitute spike noise in the k-space dataset based on the acquired k-space dataset via a trained deep learning network, the mask corresponding to a location of the spike noise in the acquired k-space dataset. An image reconstructed based on the acquired k-space dataset and the mask may be displayed.

First claim

The invention claimed is:

1. A method for magnetic resonance imaging (MRI), comprising:
acquiring a k-space dataset;
identifying, by a trained deep learning network, a set of data points which constitute spike noise in the k-space dataset;
updating the k-space dataset by removing the spike noise from the k-space dataset; and
reconstructing a magnetic resonance (MR) image from the updated k-space dataset.

2. The method of claim 1 , further comprising indicating hardware failure responsive to an area of the spike noise greater than a threshold area.

3. The method of claim 1 , further comprising acquiring another k-space dataset after displaying the reconstructed MR image, generating another set of data points which constitute spike noise in the another k-space dataset based on the another k-space dataset via the trained deep learning network, and indicating hardware failure responsive to locations of the another set of data points being the same as locations of the set of data points in the k-space dataset.

4. The method of claim 1 , wherein the trained deep learning network is trained for a particular sampling sequence that the k-space dataset is acquired, the trained deep learning network being a residual neural network.

5. The method of claim 1 , further comprising generating parameters of the trained deep learning network based on a plurality of k-space datasets and labeled noise locations in the plurality of k-space datasets.

6. The method of claim 1 , wherein the updated k-space dataset is a product of the k-space dataset and a mask showing locations of the data points which constitute spike noise in the k-space dataset.

7. The method of claim 1 , wherein the reconstruction of the MR image after removal of the spike noise is carried out upon receipt of a request to remove spike noise.

8. The method of claim 7 , wherein the request is in response to a higher than threshold spike noise in the k-space dataset or the reconstructed MR image.

9. A magnetic resonance imaging (MRI) apparatus, comprising:
a gradient coil assembly;
a radio frequency (RF) coil assembly; and
a controller coupled to the gradient coil assembly and the RF coil assembly and configured to:
acquire a k-space dataset via the gradient coil assembly and the RF coil assembly;
use a trained deep learning network to identify a set of data points which constitute spike noise in the k-space dataset;
update the k-space dataset by removing the spike noise from the acquired k-space dataset; and
reconstruct a magnetic resonance (MR) image from the updated k-space dataset,
wherein the reconstructed MR image is generated upon receipt of a request to remove spike noise.

10. The MRI apparatus of claim 9 , further comprising a memory storing the trained deep learning network.

11. The MRI apparatus of claim 9 , further comprising a display unit configured to display the reconstructed MR image.

12. The MRI apparatus of claim 9 , wherein the updated k-space dataset is a product of the acquired k-space dataset and a mask showing locations of the set of data points which constitute spike noise in the acquired k-space dataset.

13. The MRI apparatus of claim 9 , further comprising executable instructions that, when executed, cause the controller to indicate hardware failure responsive to an area of the spike noise greater than a threshold area.

14. A non-transitory computer-readable medium comprising instructions that, when executed, cause a processor to:
acquire a k-space dataset;
immediately after acquiring the k-space dataset and prior to displaying an image reconstructed based on the k-space dataset, determine, via a trained deep learning network, locations of data points that constitute noise in the k-space dataset;
update the k-space dataset by removing noise from the acquired k-space dataset;
display a magnetic resonance (MR) image reconstructed from the updated k-space dataset; and
indicate hardware failure responsive to an area of the noise greater than a threshold area.

15. The non-transitory computer-readable medium of claim 14 , wherein the trained deep learning network is trained based on a plurality of k-space datasets different from the acquired k-space dataset and labeled noise locations in the plurality of k-space dataset, the trained deep learning network being a convolutional decoder-encoder network with a number of layers greater than 100.

16. The non-transitory computer-readable medium of claim 14 , further comprising instructions that, when executed, cause the processor to normalize the k-space dataset before determining a location of the noise in the k-space dataset.

17. The non-transitory computer-readable medium of claim 14 , wherein the updated k-space dataset is generated from the acquired k-space dataset and the locations of the data points that constitute noise.

18. The non-transitory computer-readable medium of claim 14 , wherein the reconstruction of the MR image is carried out upon receipt of a request to remove spike noise, the request received in response to a higher than threshold spike noise in the acquired k-space dataset.

Assignees

• Ge Prec Healthcare Llc

Inventors

• Wu Dan
• Li Shiyu
• Dai Qingyu
• Wang Kun
• Yao Jiabin
• Lai Yongchuan

Classifications

• G06T12/20

  Inverse problem, i.e. transformations from projection space into object space · CPC title

• G01R33/565 Primary

  Correction of image distortions, e.g. due to magnetic field inhomogeneities · CPC title

• A61B2576/026

  for the brain · CPC title

• A61B5/7267

  involving training the classification device · CPC title

• A61B5/725

  using specific filters therefor, e.g. Kalman or adaptive filters (specific diagnostics methods using using bioelectric or biomagnetic signals A61B5/316) · CPC title