Back to results
US 11,696,700 B2Granted
System and method for correcting for patient motion during MR scanning
AssigneeGeneral Electric Company
InventorsMichael Rotman, Rafael Shmuel Brada, Sangtae Ahn, Christopher Judson Hardy, Itzik Malkiel, Ron Wein
Life of this patent
Estimating the term…
Abstract
K-space data obtained from a magnetic resonance imaging scan where motion was detected is split into two parts in accordance with the timing of the motion to produce first and second sets of k-space data corresponding to different poses. Sub-images are reconstructed from the k first and second sets of k-space data, which are used as inputs to a deep neural network which transforms them into a motion-corrected image.
Claims21
A magnetic resonance imaging (MRI) method comprising:
- dividing k-space data from a motion-corrupted Mill scan into a first portion and a second portion based on timing associated with transition between poses, and a predetermined scan order, wherein the predetermined scan order defines how k-space is filled as a function of time step for the motion-corrupted MM scan, wherein the first portion includes a central region of k-space data;
- reconstructing a first sub-image and a second sub-image from the first portion and the second portion, respectively;
- providing the first sub-image and the second sub-image as inputs to a neural network, wherein the neural network includes a plurality of iteration blocks and the first sub-image and the second sub-image are provided as inputs to a first iteration block of the neural network;
- and
- transforming the first sub-image and the second sub-image into a motion-corrected image using the neural network.
2 more elements
A magnetic resonance imaging (MRI) method comprising:
- producing a first sub-image and a second sub-image from a motion-corrupted set of k-space data representing a plurality of poses for an imaged subject via a process comprising:
- identifying a portion of the motion-corrupted set of k-space data corresponding to a dominant pose;
- generating the first sub-image from the identified portion of the motion-corrupted set of k-space data;
- grouping remaining portions of the motion-corrupted set of k-space data corresponding to other poses together;
- and
- generating the second sub-image from the remaining grouped portions of the motion-corrupted set of k-space data;
- inputting the first sub-image and the second sub-image to a neural network trained to transform pairs of motion-corrupted sub-images into motion-corrected images, wherein the neural network includes a plurality of iteration blocks and the first sub-image and the second sub-image are provided as inputs to a first iteration block of the neural network;
- transforming the first sub-image and the second sub-image into a motion-corrected image using the neural network;
- and
- wherein the portion of the motion-corrupted set of k-space data corresponding to a dominant pose includes a central region of the motion-corrupted set of k-space data.
7 more elements
A magnetic resonance imaging (MRI) reconstruction module comprising:
- circuitry programmed to implement a trained neural network for transformation of motion-corrupted images into motion-corrected images;
- a plurality of iteration blocks of the trained neural network, wherein the trained neural network receives a first sub-image corresponding to a first pose and a second sub-image corresponding to at least a second pose as inputs, and wherein the trained neural network outputs a motion-corrected image;
- a data consistency block of each iteration block of the plurality of iteration blocks, wherein the data consistency block is programmed to produce a reconstructed image where line locations in k-space that correspond to the first pose are overwritten with k-space data from the first pose;
- wherein the first pose includes a pre-motion state, and the second pose includes a post-motion state;
- and
- wherein the trained neural network includes a plurality of iteration blocks and the first sub-image and the second sub-image are provided as inputs to a first iteration block of the trained neural network.
3 more elements
Dependencies are inferred from the claim text.