Methods, apparatuses, systems and computer-readable mediums for correcting echo planar imaging artifacts

1 . A method of correcting echo planar imaging artifacts, the method comprising:

correcting, using a trained neural network, at least one echo planar imaging artifact in an image to obtain a corrected image, wherein

the trained neural network is trained using a dataset of artifacted echo planar images,

the dataset of artifacted echo planar images includes high-resolution images that have been generated by an imaging process other than echo planar imaging and modified to include at least one first simulated echo planar imaging artifact, and

the at least one first simulated echo planar imaging artifact is at least one of a simulated B0 susceptibility artifact, a simulated chemical shift artifact, a simulated Rician noise artifact, a simulated Nyquist artifact, or a simulated Gibbs ringing artifact.

2 . The method of claim 1 , wherein the image is obtained through echo planar imaging.

3 . The method of claim 1 , wherein the dataset of artifacted echo planar images further includes at least one artifacted image with at least one second simulated echo planar imaging artifact.

4 . The method of claim 1 , wherein the dataset of artifacted echo planar images further includes at least one artifacted image with at least one deliberately induced echo planar imaging artifact generated by altering imaging equipment.

5 . The method of claim 1 , wherein the at least one echo planar imaging artifact includes at least one of a B0 susceptibility artifact, a chemical shift artifact, a Rician noise artifact, a Nyquist artifact, or a Gibbs ringing artifact.

6 . A method of training a neural network to correct single-shot echo planar imaging artifacts, the method comprising:

inputting a dataset of artifacted single-shot echo planar images into a neural network, each artifacted echo planar image including at least one synthetic echo planar imaging artifact;

modifying, via the neural network, each image of the dataset of artifacted single-shot echo planar images to remove the at least one synthetic echo planar imaging artifact; and

updating one or more parameters of the neural network based on the modified images of the dataset of artifacted single-shot echo planar images,

wherein the dataset of artifacted single-shot echo planar images includes high-resolution images that have been generated by an imaging process other than echo planar imaging and modified to include the at least one synthetic echo planar imaging artifact and the at least one synthetic echo planar imaging artifact is at least one of a simulated B0 susceptibility artifact, a simulated chemical shift artifact, a simulated Rician noise artifact, a simulated Nyquist artifact, or a simulated Gibbs ringing artifact.

7 . The method of claim 6 , further comprising:

validating the neural network with a validation dataset of artifacted single-shot echo planar images, each artifacted single-shot echo planar image, in the validation dataset of artifacted single-shot echo planar images, including at least one synthetic echo planar imaging artifact.

8 . The method of claim 6 , wherein the at least one synthetic echo planar imaging artifact includes at least one of a B0 susceptibility artifact, a chemical shift artifact, a Rician noise artifact, a Nyquist artifact, or a Gibbs ringing artifact.

9 . The method of claim 6 , wherein the neural network is trained using a combination of supervised and unsupervised learning.

10 . The method of claim 6 , wherein the neural network employs at least one of a generative adversarial network (GAN) multiscale generator model architecture, a GAN multiscale generator with attention architecture, a GAN multiscale generator model with attention architecture, or a GAN generator model with attention and with recurrent convolutional layers architecture.

11 . The method of claim 10 , wherein the neural network is trained to minimize at least one of a discriminative loss, a Wasserstein loss, or a loss computed using a pretrained model.

12 . The method of claim 6 , wherein the neural network is trained with an iterative fine-tuning approach.