Retrospective motion correction using a combined neural network and model-based image reconstruction of magnetic resonance data
AssigneeThe General Hospital Corporation
InventorsStephen Cauley, Melissa Haskell, Lawrence Wald
Life of this patent
Estimating the term…
Abstract
A combined physics-based and machine learning framework is used for reconstructing images from k-space data, in which motion artifacts are significantly reduced in the reconstructed images. In general, model-based retrospective motion correction techniques are accelerated using fast machine learning (“ML”) steps, which may be implemented using a trained neural network such as a convolutional neural network. In this way, the confidence of a classical physics-based reconstruction is obtained with the computational benefits of an ML-based network.
Claims17
A method for reconstructing an image with reduced motion artifacts from k-space data acquired with a magnetic resonance imaging (MRI) system, the method comprising:
- (a) accessing k-space data with a computer system, the k-space data having been acquired from a subject using a magnetic resonance imaging (MRI) system, wherein the k-space data contain motion errors associated with motion of the subject;
- (b) reconstructing a motion-corrupted image of the subject from the k-space data using the computer system;
- (c) reconstructing a motion-corrected image of the subject from the k-space data using the computer system by:
- inputting the motion-corrupted image to a trained neural network, generating output as a motion artifact image that depicts motion artifacts associated with the motion of the subject;
- generating a difference image by subtracting the motion artifact image from the motion-corrupted image;
- computing motion parameters that define one or more motion trajectories from the difference image and the k-space data;
- and
- reconstructing the motion-corrected image of the subject by inputting the k-space data and the motion parameters to a model-based image reconstruction algorithm, generating output as the motion-corrected image.
Dependencies are inferred from the claim text.