K-space trajectory infidelity correction in magnetic resonance imaging
For k-space trajectory infidelity correction, a model is machine trained to correct k-space measurements in k-space. K-space trajectory infidelity correction uses deep learning. Trajectory infidelity is corrected from a k-space point of view. Since the image artifacts arise from k-space acquisition distortion, a machine learning model is trained to correct in k-space, either changing values of k-space measurements or estimating the trajectory shifts in k-space.
1. A method for k-space trajectory infidelity correction in a magnetic resonance (MR) imaging system, the method comprising:
scanning, by the MR imaging system, a patient with an MR sequence, the scanning resulting in k-space measurements corresponding to k-space trajectories;
correcting the k-space measurements for errors in the k-space trajectories, the correcting being by input of the k-space measurements to a machine-learned model which outputs corrected k-space measurements in response to the input, the machine-learned model having been trained using a loss for trajectory shift, a loss for k-space correction, and a loss for an estimate of corrupted data estimated from the trajectory shift and the k-space correction;
reconstructing an MR image from the corrected k-space measurements; and
displaying the MR image.
2. The method of claim 1 wherein scanning comprises echo planar imaging.
3. The method of claim 1 wherein scanning comprises scanning with the k-space trajectories comprising non-Cartesian trajectories.
4. The method of claim 1 wherein correcting comprises correcting by the machine-learned model, the machine-learned model comprising a deep learned autoencoder network.
5. The method of claim 1 wherein reconstructing comprises reconstructing a two- or three-dimensional distribution of pixels or voxels, respectively, representing an area or volume, respectively, of the patient, and wherein displaying comprises rendering from the voxels or pixels to a two-dimensional display.
6. The method of claim 1 wherein correcting comprises correcting by the machine-learned model having been trained as a first neural network trained in conjunction with a second neural network trained to estimate trajectory shifts.
7. The method of claim 1 further comprising estimating trajectory shifts by another machine-learned model, the trajectory shifts estimated in response to input of the k-space data from the scanning to the other machine-learned model.
8. A system for trajectory correction in magnetic resonance (MR) imaging, the system comprising:
an MR scanner configured to scan a patient, the scan providing first scan data in a scan domain;
an image processor configured to alter trajectories of the first scan data by application of a machine-learned model to the first scan data, the machine-learned model outputting second scan data in the scan domain, the second scan data having the altered trajectories, and the image processor configured to reconstruct a representation in an object domain from the second scan data in the scan domain, wherein the machine-learned model was previously trained using a loss for trajectory shift, a loss for k-space correction, and a loss for an estimate of corrupted data estimated from the trajectory shift and the k-space correction; and
a display configured to display an MR image from the reconstructed representation.
9. The system of claim 8 wherein the MR scanner is configured to scan the patient with echo planar imaging or non-Cartesian trajectories.
10. The system of claim 8 wherein the machine-learned model comprises a deep learned encoder-decoder network.
11. The system of claim 8 wherein the representation is a three-dimensional distribution of voxels representing a volume of the patient, and wherein the MR image is a rendering of the voxels to the display comprising a two-dimensional display.
12. The system of claim 8 wherein the machine-learned model was trained as a first neural network trained in conjunction with a second neural network trained to estimate trajectory shifts.
13. The system of claim 8 wherein the image processor is configured to estimate trajectory shifts by application of another machine-learned model, the trajectory shifts estimated in response to input of the first scan data in the scan domain to the other machine-learned model.
14. A method for training a network for reducing artifacts from trajectory infidelity in magnetic resonance (MR) imaging, the method comprising:
defining a first neural network to receive input k-space data;
defining a second neural network to receive the input k-space data and output a trajectory shift;
jointly machine training the first neural network and second neural network for correction of trajectory infidelity in the input k-space data; and
storing a machine-learned model resulting from the joint training.
15. The method of claim 14 wherein defining comprises defining the first neural network as an image-to-image network.
16. The method of claim 14 wherein machine training comprises deep learning by the first neural network to output changes to trajectories of the input k-space data in response to input of the input k-space data;
further comprising altering the trajectories of the input k-space data according to the output changes; and
wherein machine training further comprises machine training with a loss based on a difference between the input k-space data with the altered trajectories and a ground truth.
17. The method of claim 14 wherein machine training comprises the joint training with first, second and third losses, the first loss being for corrected k-space data, the second loss being for the trajectory shift, and the third loss being for corrupted k-space data formed from the corrected k-space data and the trajectory shift.