System and method for reconstructing ghost-free images from data acquired using simultaneous multislice magnetic resonance imaging

1. A method for reconstructing images of a subject using a magnetic resonance imaging (MRI) system, the steps of the method comprising:

(a) providing to a computer system, training data acquired from a subject, wherein the training data includes first training data acquired from a plurality of different slices in the subject using a first series of alternating readout gradients, and second training data acquired from the plurality of different slices in the subject using a second series of alternating readout gradients that have opposite polarity from the first series of alternating readout gradients;

(b) generating from the training data with the computer system, ghost-free data for each of the plurality of different slices in the subject;

(c) generating with the computer system, positive readout data by combining k-space lines in the first training data acquired with a positive readout gradient and k-space lines in the second training data acquired with a positive readout gradient;

(d) generating with the computer system, negative readout data by combining k-space lines in the first training data acquired with a negative readout gradient and k-space lines in the second training data acquired with a negative readout gradient;

(e) synthesizing from the positive readout data with the computer system, first collapsed slice data that synthesizes the positive readout data from each of the plurality of different slices in the subject as having been acquired simultaneously;

(f) synthesizing from the negative readout data with the computer system, second collapsed slice data that synthesizes the negative readout data from each of the plurality of different slices in the subject as having been acquired simultaneously;

(g) computing with the computer system, a reconstruction kernel for each of the plurality of different slice locations in the subject using a training algorithm in which the first and second collapsed slice data are used as source data and the ghost-free data are used as target data;

(h) acquiring data simultaneously from the plurality of different slices in the subject using an MRI system performing a simultaneous multislice acquisition pulse sequence;

(i) reconstructing images of each of the plurality of different slices by applying the reconstruction kernels to the acquired data with the computer system, wherein the images are substantially free from ghost artifacts.

2. The method as recited in claim 1 , wherein the training data includes autocalibration signal data.

3. The method as recited in claim 1 , wherein step (b) includes generating the ghost-free data by correcting the first and second training data for phase errors in at least a k x -direction in k-space.

4. The method as recited in claim 3 , wherein step (b) includes generating the ghost-free data by correcting the first and second training data for phase errors in both the k x -direction in k-space and a k y -direction in k-space.

5. The method as recited in claim 1 , wherein the ghost-free data are generated using a ghost elimination via spatial and temporal encoding method.

6. The method as recited in claim 1 , wherein the reconstruction kernels are computed in step (g) using a slice-GRAPPA training procedure.

7. The method as recited in claim 1 , wherein step (h) includes acquiring the data using in-plane acceleration.

8. The method as recited in claim 7 , wherein the reconstruction kernels computed in step (g) account for both in-plane and through-plane acceleration, and wherein the training algorithm also uses the ghost-free data as source data.

9. A computer-implemented method for computing reconstruction kernels for reconstructing images from magnetic resonance data simultaneously acquired from a plurality of different slices in a subject using a magnetic resonance imaging (MRI) system, the steps of the method comprising:

(a) acquiring first training data from the plurality of different slices in the subject using the MRI system performing a first pulse sequence having a first series of alternating readout gradients;

(b) acquiring second training data from the plurality of different slices in the subject using the MRI system performing a second pulse sequence having a second series of alternating readout gradients that have opposite polarity from the first series of alternating readout gradients;

(c) generating from the first and second training data with a computer system, ghost-free data for each of the plurality of different slices in the subject;

(d) synthesizing from the first and second training data with the computer system, collapsed slice data that synthesizes the first and second training data from each of the plurality of different slices in the subject as having been acquired simultaneously;

(g) computing with the computer system, a reconstruction kernel for each of the plurality of different slice locations in the subject using a training algorithm in which the collapsed slice data are used as source data and the ghost-free data are used as target data; and

(h) storing the reconstruction kernels in the computer system.

10. The method as recited in claim 9 , wherein the first and second training data include autocalibration signal data.

11. The method as recited in claim 9 , wherein step (c) includes generating the ghost-free data by correcting the first and second training data for phase errors in at least a k x -direction in k-space.

12. The method as recited in claim 11 , wherein step (c) includes generating the ghost-free data by correcting the first and second training data for phase errors in both the k x -direction in k-space and a k y -direction in k-space.

13. The method as recited in claim 9 , wherein the ghost-free data are generated using a ghost elimination via spatial and temporal encoding method.

14. The method as recited in claim 9 , wherein the reconstruction kernels are computed in step (g) using a slice-GRAPPA training procedure.

15. The method as recited in claim 9 , wherein the reconstruction kernels computed in step (g) account for both in-plane and through-plane acceleration, and wherein the training algorithm also uses the ghost-free data as source data.

16. The method as recited in claim 9 , wherein step (d) includes:

generating with the computer system, positive readout data by combining k-space lines in the first training data acquired with a positive readout gradient and k-space lines in the second training data acquired with a positive readout gradient;

generating with the computer system, negative readout data by combining k-space lines in the first training data acquired with a negative readout gradient and k-space lines in the second training data acquired with a negative readout gradient;

synthesizing from the positive readout data with the computer system, first collapsed slice data that synthesizes the positive readout data from each of the plurality of different slices in the subject as having been acquired simultaneously;

synthesizing from the negative readout data with the computer system, second collapsed slice data that synthesizes the negative readout data from each of the plurality of different slices in the subject as having been acquired simultaneously; and

wherein the collapsed slice data comprises the first and second collapsed slice data.