Auto-calibrating parallel MRI technique with distortion-optimal image reconstruction
View Patent ↗The invention is a new computational method for the formation of magnetic resonance (MR) images. The method utilizes the data acquired by the multiple receiver channels available as parallel imaging hardware on standard MRI scanners to: (i) automatically identify a set of multi-input multi-output (MIMO) systems (e.g., MIMO filter banks) that act as interpolation kernels for acquired MR data sets (that can be subsampled with respect to the Nyquist criterion) without requiring a separate calibration scan; and (ii) use the identified MIMO systems to synthesize MR data sets that can in turn be used to produce high quality images, thereby enabling high quality imaging with fewer data samples than current methods (or equivalently provide higher image quality with the same number of data samples). A unique feature of the present invention is its ability to account for aliasing effects and minimize the associated image distortion by optimally adapting the said MIMO interpolation (image reconstruction) kernels. This ability to image with a reduced number of data samples accelerates the imaging process; hence, overcoming the main shortcoming of MRI compared to other medical imaging modalities.
1. A method for acquiring and producing images of a subject with a magnetic resonance imaging (MRI) system equipped with a plurality of receiver coils, the method comprising the steps of:
(a) acquiring a plurality of MR data sets in k-space from at least two of said receiver coils;
(b) computing at least one interpolation kernel that optimizes a cost function comprising:
(i) a measure of data consistency; and
(ii) a measure of predicted image distortions corresponding to aliasing effects;
with both measures computed with respect to at least one of said MR data sets acquired in step (a);
(c) performing interpolation of said MR data sets from step (a) using said one or more interpolation kernels from step (b) in k-space or via point-by-point multiplication in image space; and
(d) reconstructing a plurality of images from the interpolated data sets produced in step (c).
2. The method as recited in claim 1 , wherein the acquired k-space data sets may include auto-calibration scan (ACS) data sets according to the method of GRAPPA or other coil calibration data.
3. The method as recited in claim 2 wherein step (c) or step (d) also utilize at least one sample value of said ACS or coil calibration data.
4. The method as recited in claim 1 , wherein each acquired data set in step (a) is sampled on a Cartesian k-space trajectory.
5. The method as recited in claim 1 , wherein the spacing of data samples in at least one of the data sets acquired in step (a) is subsampled with respect to the Nyquist sampling criterion for a prescribed field-of-view (FOV); and the intermediate data sets synthesized in step (c) correspond to a plurality of data sets that are Nyquist-sampled with respect to said prescribed FOV.
6. The method as recited in claim 1 wherein an estimate of receiver coil noise characteristics or correlation of noise among the receiver coils is utilized in computation of the interpolation kernel in step (b) or computations in steps (c) or (d).
7. The method as recited in claim 1 , further comprising the step of combining the plurality of images reconstructed in step (d) into a composite image.
8. A method for acquiring and producing images of a subject with a MRI system equipped with a plurality of receiver coils, the method comprising the steps of:
(a) acquiring a plurality of MR data sets in k-space from at least two of said receiver coils;
(b) identifying at least one multi-input filter bank that is comprised of a plurality of finite impulse response (FIR) filters and optimizes a cost function comprising:
(i) a measure of data consistency; and
(ii) a measure of predicted image distortions corresponding to aliasing effects;
with both measures computed with respect to at least one of said MR data sets;
(c) synthesizing at least one intermediate data set from said MR data sets from step (a) using said filter banks from step (b); and
(d) reconstructing at least one image from the intermediate data sets synthesized in step (c).
9. The method as recited in claim 8 , wherein said intermediate data sets in step (c) are computed as outputs of said filter bank with inputs comprising said MR data sets from step (a).
10. The method as recited in claim 8 , wherein the acquired k-space data sets may include auto-calibration scan (ACS) data sets according to the method of GRAPPA or other coil calibration data.
11. The method as recited in claim 10 wherein step (c) or step (d) also utilize at least one sample value of said ACS or coil calibration data.
12. The method as recited in claim 8 , wherein each acquired data set in step (a) is sampled on a Cartesian k-space trajectory.
13. The method as recited in claim 8 , wherein the spacing of data samples in at least one of the data sets acquired in step (a) is subsampled with respect to the Nyquist sampling criterion for a prescribed field-of-view (FOV); and the intermediate data sets synthesized in step (c) correspond to a plurality of data sets that are Nyquist-sampled with respect to said prescribed FOV.
14. The method as recited in claim 8 , wherein an estimate of receiver coil noise characteristics or correlation of noise among the receiver coils is utilized in identification of the multi-input filter bank in step (b) or computations in steps (c) or (d).
15. The method of as recited in claim 8 , further comprising the step of combining the plurality of images reconstructed in step (d) into a composite image.
16. The method as recited in claim 8 , wherein each of the k-space data sets in step (a) is acquired by two-dimensional Cartesian acquisition and the computation in step (c) is performed using discrete Fourier transform operators in the image space or in a hybrid space.
17. A method for acquiring and producing images of a subject with a MRI system equipped with a plurality of receiver coils, the method comprising the steps of:
(a) acquiring a plurality of MR data sets in k-space from at least two of said receiver coils;
(b) identifying at least one multi-input filter bank that is comprised of a plurality of shift-invariant filters and optimizes a cost function comprising:
(i) a measure of data consistency; and
(ii) a measure of predicted image distortions corresponding to aliasing effects;
with both measures computed with respect to at least one of said MR data sets;
(c) synthesizing at least one intermediate data set from said MR data sets from step (a) using said filter banks from step (b); and
(d) reconstructing at least one image from the intermediate data sets synthesized in step (c).
18. The method as recited in claim 17 , wherein said intermediate data sets in step (c) are computed as outputs of said filter bank with inputs comprising said MR data sets from step (a).
19. The method as recited in claim 17 , wherein the acquired k-space data sets may include auto-calibration scan (ACS) data sets according to the method of GRAPPA or other coil calibration data.
20. The method as recited in claim 19 , wherein step (c) or step (d) also utilize at least one sample value of said ACS or coil calibration data.
21. The method as recited in claim 17 , wherein each acquired data set in step (a) is sampled on a Cartesian k-space trajectory.
22. The method as recited in claims 17 , wherein the spacing of data samples in at least one of the data sets acquired in step (a) corresponds to subsampling with respect to the Nyquist sampling criterion for a prescribed field-of-view (FOV); and the intermediate data sets synthesized in step (c) correspond to a plurality of data sets that are Nyquist-sampled with respect to said prescribed FOV.
23. The method as recited in claim 17 , wherein an estimate of receiver coil noise characteristics or correlation of noise among the receiver coils is utilized in identification of the multi-input filter bank in step (b) or computations in steps (c) or (d).
24. The method of as recited in claim 17 further comprising the step of combining the plurality of images reconstructed in step (d) into a composite image.
25. The method as recited in claim 17 , wherein each of the k-space data sets in step (a) is acquired by two-dimensional Cartesian acquisition and the computation in step (c) is performed using discrete Fourier transform operators in the image space or in a hybrid space.
26. The method as recited in claim 17 , in which said measures of data consistency and predicted image distortion in step (b) are computed as functions of the discrepancy of the left and right hand sides of the “blind PI relations.”
27. The method as recited in claim 17 , wherein said multi input filter bank consists of FIR filters and said cost function is the “ACSIOM cost function” or any version thereof corresponding to linear transformations of the arguments.