MRI reconstruction with image domain optimization

1. An apparatus for reconstructing a magnetic resonance imaging (MRI) image, the apparatus comprising a processor and a memory storing executable instructions, that in response to execution by the processor, cause the apparatus to:

access a training set of MRI data;

build a discriminator model using the training set of MRI data, the discriminator model being trained to distinguish between fully sampled MRI data and MRI data that comprises undersampled MRI data or reconstructed MRI data that is not fully sampled MRI data;

during inference, use the discriminator model to classify MRI data from an output of an MRI data optimization model as fully sampled MRI data; and

if the discriminator model classifies the MRI data as fully sampled MRI data, generate a reconstructed MRI image from the fully sampled MRI data; or

if the discriminator model does not classify the MRI data from MRI data optimization model as fully sampled MRI data, continue processing the MRI data with the MRI data optimization model until the discriminator model classifies the MRI data as fully sampled MRI data.

2. The apparatus according to claim 1 , wherein the MRI data optimization model is configured to optimize the undersampled MRI data or the reconstructed MRI data that is not fully sampled MRI data until the discriminator classifies the MRI data as fully sampled MRI data to generate the reconstructed MRI image.

3. The apparatus according to claim 1 wherein the training set of MRI data comprises paired MRI data or unpaired MRI data.

4. The apparatus according to claim 1 wherein the apparatus is further configured to:

receive undersampled MRI data or reconstructed MRI data from an MRI apparatus;

optimize parameters of the undersampled MRI data or reconstructed MRI data using the MRI data optimization model until the discriminator classifies the output of the MRI data optimization model as fully sampled image data; and

generate the reconstructed MRI image from the output of the MRI data optimization model when the discriminator classifies the output of the MRI data optimization model as fully sampled MRI data.

5. The apparatus according to claim 4 , wherein the undersampled MRI data is directly optimized.

6. The apparatus according to claim 5 , wherein the undersampled MRI data is directly optimized by changing values in image data or k-space using a neural network.

7. The apparatus according to claim 1 , wherein during training the discriminator model and the MRI data optimization model are trained alternatively.

8. The apparatus according to claim 1 wherein the MRI data comprises MRI images or MRI k-space data.

9. The apparatus according to claim 1 , wherein during inference, parameters of the discriminator model are fixed.

10. A method for reconstructing a magnetic resonance imaging (MRI) image, the method comprising:

using a hardware processor to access a training set of MRI data;

using a hardware processor to build a discriminator model using the training set of MRI data, the discriminative model being trained to distinguish between fully sampled MRI data and MRI data that comprises undersampled MRI data or reconstructed MRI data that is not fully sampled MRI data;

during inference, use the discriminator model to classify MRI data from an output of an MRI data optimization model as fully sampled MRI data; and

if the discriminator model classifies the MRI data from the MRI data optimization model as fully sampled MRI data, generate a reconstructed MRI image from the MRI data; or

if the discriminator model does not classify the MRI data from the MRI data optimization model as fully sampled MRI data, continue processing the MRI data with the MRI data optimization model until the discriminator model classifies the MRI data as fully sampled MRI data.

11. The method according to claim 10 , further comprising using the MRI data optimization model to optimize the undersampled MRI data or the reconstructed MRI data that is not fully sampled MRI data until the discriminator classifies the MRI data as fully sampled MRI data to generate the reconstructed MRI image.

12. The method according to claim 10 wherein the training set of MRI data comprises paired MRI data or unpaired MRI data.

13. The method according to claim 10 , the method further comprising:

receiving undersampled MRI data or reconstructed MRI data from an MRI apparatus;

optimizing parameters of the undersampled MRI data or reconstructed MRI data using the MRI data optimization model until the discriminator classifies the output of the MRI data optimization model as fully sampled image data; and

generating the reconstructed MRI image from the output of the MRI data optimization model when the discriminator classifies the output of the MRI data optimization model as fully sampled MRI data.

14. The method according to claim 13 , wherein the undersampled MRI data is directly optimized.

15. The method according to claim 14 , wherein the undersampled MRI data is directly optimized by changing values in image data or k-space using a neural network.

16. The method according to claim 10 , wherein during training the discriminator model and the MRI data optimization model are trained alternatively.

17. The method according to claim 10 wherein the MRI data comprises MRI images or MRI k-space data.

18. The method according to claim 10 , the method further comprising fixing parameters of the discriminator model during inference.

19. The method according to claim 10 , the method further comprising using undersampled MRI data, MRI data with random values or a blank image with zero values as initial MRI data for the MRI data optimization model.

20. A computer program product comprising non-transitory machine readable instruction, which when executed are configured to carry out the method according to claim 10 .