Systems and methods of reconstructing magnetic resonance images using deep learning

A magnetic resonance (MR) image reconstruction system comprising an MR image reconstruction computing device, the MR image reconstruction computing device comprising at least one processor in communication with at least one memory device, and the at least one processor programmed to:

• receive MR signals of a subject;
• reconstruct crude MR images based on the MR signals;
• analyze the crude MR images using a neural network model, wherein the neural network model was trained by:
• inputting a first subset of training MR images to the neural network model;
• and
• setting a second subset of the training MR images as target outputs of the neural network model, wherein each image in the first subset is acquired during a neighboring respiratory phase of at least one of the images in the second subset, the first subset and the second subset acquired with the same imaging parameters
• wherein the neural network model is trained to retain signals across respiratory phases and reduce artifacts in input images to the neural network model;
• derive clear MR images based on the analysis, wherein the clear MR images include reduced artifacts, compared to the crude MR images;
• and
• output the clear MR images.

A computer-implemented method of reducing artifacts in magnetic resonance (MR) images of a subject, comprising:

• receiving crude MR images reconstructed based on MR signals;
• and
• training a neural network model using the crude MR images as training MR images by:
• inputting a first subset of the crude MR images to the neural network model;
• setting a second subset of the crude images as target outputs of the neural network model, wherein each image in the first subset was acquired during a neighboring respiratory phase of at least one of the images in the second subset, the first subset and the second subset acquired with the same imaging parameters;
• analyzing the first subset of the crude MR images using the neural network model;
• comparing outputs of the neural network model with the target outputs;
• and
• adjusting the neural network model based on the comparison,
• wherein the trained neural network model is configured to retain signals across respiratory phases and reduce artifacts in the crude MR images.