US 12,118,720 B2
Systems and methods of magnetic resonance image processing using neural networks having reduced dimensionality
Robert Marc Lebel, Calgary (CA); Suryanarayanan S. Kaushik, Wauwatosa, WI (US); Graeme C. Mckinnon, Hartland, WI (US); and Xucheng Zhu, Mountain View, CA (US)
Assigned to GE PRECISION HEALTHCARE LLC, Wauwatosa, WI (US)
Filed by GE PRECISION HEALTHCARE LLC, Wauwatosa, WI (US)
Filed on Dec. 17, 2021, as Appl. No. 17/644,857.
Prior Publication US 2023/0196556 A1, Jun. 22, 2023
Int. Cl. G06T 7/00 (2017.01); G06N 3/04 (2023.01); G06N 3/045 (2023.01); G06N 3/08 (2023.01); G16H 30/40 (2018.01)
CPC G06T 7/0012 (2013.01) [G06N 3/045 (2023.01); G06N 3/08 (2013.01); G16H 30/40 (2018.01); G06T 2207/10088 (2013.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01)] 19 Claims
OG exemplary drawing
 
1. A magnetic resonance (MR) image processing system, comprising an MR image processing computing device, the MR image processing computing device comprising at least one processor in communication with at least one memory device, and the at least one processor programmed to:
execute a neural network model, wherein the neural network model includes a plurality of neural networks and is configured to receive crude MR data having a first number of dimensions as an input and configured to output processed MR images associated with the crude MR data, the processed MR images having the first number of dimensions;
receive a pair of pristine data and corrupted data, wherein the pristine data and the corrupted data have a second number of dimensions that is lower than the first number of dimensions, and the corrupted data are the pristine data added with primitive features; and
train the neural network model using the pair of the pristine data and the corrupted data by:
inputting the corrupted data to the neural network model;
setting the pristine data as target outputs of the neural network model;
analyzing the corrupted data 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 change primitive features associated with the crude MR data.