System and method for detecting motion-ridden shots in multi-shot acquisitions and utilizing deep learning based reconstruction for motion correction
A method includes obtaining k-space data, wherein a plurality of navigator like echoes of the k-space data including an additional navigator like echo are acquired for each shot or a group of shots. The k-space data is motion corrupted. The method includes identifying any shots where a subject moved during acquisition based on the respective additional navigator like echoes. The method includes generating dominant pose k-space data based on identification of any shots where the subject moved during acquisition, the dominant pose k-space data includes only shots not affected by movement, wherein the dominant pose k-space data is missing k-space data due to rejecting the shots where the subject moved. The method includes utilizing a deep learning-based reconstruction model on the motion-corrupted k-space data to modify motion-corrupted k-space data with k-space data that is consistent with the dominant pose k-space data to generate a reconstructed image.
1 . A computer-implemented method, comprising:
obtaining, via a processor, k-space data of a subject acquired with a magnetic resonance imaging (MRI) scanner, wherein a portion of the k-space data is motion corrupted, and wherein a plurality of navigator like echoes of the k-space data comprising an additional navigator like echo are acquired for each shot or a group of shots of a plurality of shots for a respective slice;
identifying, via the processor, any shots of the plurality of shots where the subject moved during acquisition based on respective additional navigator like echoes for the plurality of shots;
generating, via the processor, dominant pose k-space data based on identification of any shots of the plurality of shots where the subject moved during acquisition, the dominant pose k-space data comprising only shots of the plurality of shots not affected by movement of the subject during acquisition, wherein the dominant pose k-space data is missing k-space data due to rejecting the shots where the subject moved during acquisition; and
utilizing, via the processor, a deep learning-based reconstruction model on the k-space data to modify the portion of the motion-corrupted k-space data that is motion corrupted with other k-space data that is consistent with the dominant pose k-space data to generate a reconstructed image from the k-space data, wherein utilizing the deep learning-based reconstruction model on the k-space data comprises:
transforming the portion of the k-space data that is motion corrupted to a motion-corrupted image;
transforming centrally located phase encoded lines from the k-space data into a contrast image, wherein the k-space data is derived from both a first set of shots where the subject moved during acquisition and a second set of shots not affected by movement of the subject during acquisition;
inputting both the motion-corrupted image and the contrast image into the deep learning-based reconstruction model, wherein the dominant pose k-space data is utilized for application of soft data consistency in an unrolled framework; and
outputting from the deep learning-based reconstruction model the reconstructed image.
2 . The computer-implemented method of claim 1 , wherein the additional navigator like echo is acquired at a start, a middle, or an end of each shot of the plurality of shots for the respective slice where the additional navigator like echo is acquired.
3 . The computer-implemented method of claim 1 , further comprising comparing the respective additional navigator like echoes for the plurality of shots to identify a subset of shots of the plurality of shots that corresponds to a dominant pose of the subject and to identify any shots of the plurality of shots where the subject moved during acquisition.
4 . The computer-implemented method of claim 3 , wherein comparing the respective additional navigator like echoes comprises clustering the respective additional navigator like echoes to obtain positioning information to predict any shot where the subject moved during acquisition.
5 . The computer-implemented method of claim 1 , further comprising:
generating, via the processor, a k-space mask from the k-space data utilizing only the shots not affected by movement of the subject during acquisition, wherein the k-space mask is configured to mask the shots where the subject moved during acquisition; and
applying, via the processor, the k-space mask to the k-space data to generate the dominant pose k-space data.
6 . A system, comprising:
a memory encoding processor-executable routines; and
a processor configured to access the memory and to execute the processor-executable routines, wherein the processor-executable routines, when executed by the processor, cause the processor to:
obtain k-space data of a subject acquired with a magnetic resonance imaging (MRI) scanner, wherein a portion of the k-space data is motion corrupted, and wherein a plurality of navigator like echoes of the k-space data comprising an additional navigator like echo are acquired for each shot or a group of shots of a plurality of shots for a respective slice;
identify any shots of the plurality of shots where the subject moved during acquisition based on respective additional navigator like echoes for the plurality of shots;
generate dominant pose k-space data based on identification of any shots of the plurality of shots where the subject moved during acquisition, the dominant pose k-space data comprising only shots of the plurality of shots not affected by movement of the subject during acquisition, wherein the dominant pose k-space data is missing k-space data due to rejecting the shots where the subject moved during acquisition; and
utilize a deep learning-based reconstruction model on the k-space data to modify the portion of the k-space data that is motion corrupted with other k-space data that is consistent with the dominant pose k-space data to generate a reconstructed image from the k-space data, wherein utilizing the deep learning-based reconstruction model on the k-space data comprises:
transforming the portion of k-space data that is motion corrupted to a motion-corrupted image;
transforming centrally located phase encoded lines from the k-space data into a contrast image, wherein the k-space data is derived from both a first set of shots where the subject moved during acquisition and a second set of shots not affected by movement of the subject during acquisition;
inputting both the motion-corrupted image and the contrast image into the deep learning-based reconstruction model, wherein the dominant pose k-space data is utilized for application of soft data consistency in an unrolled framework; and
outputting from the deep learning-based reconstruction model the reconstructed image.
7 . The system of claim 6 , wherein the additional navigator like echo is acquired at a start, a middle, or an end of each shot of the plurality of shots for the respective slice where the additional navigator like echo is acquired.
8 . The system of claim 6 , wherein the processor-executable routines, when executed by the processor, further cause the processor to compare the respective additional navigator like echoes for the plurality of shots to identify a subset of shots of the plurality of shots that corresponds to a dominant pose of the subject and to identify any shots of the plurality of shots where the subject moved during acquisition.
9 . The system of claim 8 , wherein comparing the respective additional navigator like echoes comprises clustering the respective additional navigator like echoes to obtain positioning information to predict any shot where the subject moved during acquisition.
10 . The system of claim 6 , wherein the processor-executable routines, when executed by the processor, further cause the processor to:
generate a k-space mask from the k-space data utilizing only the shots not affected by movement of the subject during acquisition, wherein the k-space mask is configured to mask the shots where the subject moved during acquisition; and
apply the k-space mask to the k-space data to generate the dominant pose k-space data.
11 . A non-transitory computer-readable medium, the non-transitory computer-readable medium comprising processor-executable code that when executed by a processor, causes the processor to:
obtain k-space data of a subject acquired with a magnetic resonance imaging (MRI) scanner, wherein a portion of the k-space data is motion corrupted, and wherein a plurality of navigator like echoes of the k-space data comprising an additional navigator like echo are acquired for each shot or a group of shots of a plurality of shots for a respective slice;
identify any shots of the plurality of shots where the subject moved during acquisition based on respective additional navigator like echoes for the plurality of shots;
generate dominant pose k-space data based on identification of any shots of the plurality of shots where the subject moved during acquisition, the dominant pose k-space data comprises only shots of the plurality of shots not affected by movement of the subject during acquisition, wherein the dominant pose k-space data is missing k-space data due to rejecting the shots where the subject moved during acquisition; and
utilize a deep learning-based reconstruction model on the k-space data to modify the portion of the k-space data that is motion corrupted with other k-space data that is consistent with the dominant pose k-space data to generate a reconstructed image from the k-space data, wherein utilizing the deep learning-based reconstruction model on the k-space data comprises:
transforming the portion of the k-space data that is motion corrupted to a motion-corrupted image;
transforming centrally located phase encoded lines from the k-space data into a contrast image, wherein the k-space data is derived from both a first set of shots where the subject moved during acquisition and a second set of shots not affected by movement of the subject during acquisition;
inputting both the motion-corrupted image and the contrast image into the deep learning-based reconstruction model, wherein the dominant pose k-space data is utilized for application of soft data consistency in an unrolled framework; and
outputting from the deep learning-based reconstruction model the reconstructed image.
12 . The non-transitory computer-readable medium of claim 11 , wherein the additional navigator like echo is acquired at a start, a middle, or an end of each shot of the plurality of shots for the respective slice where the additional navigator like echo is acquired.
13 . The non-transitory computer-readable medium of claim 11 , wherein the processor-executable code, when executed by the processor, further cause the processor to compare the respective additional navigator like echoes for the plurality of shots to identify a subset of shots of the plurality of shots that corresponds to a dominant pose of the subject and to identify any shots of the plurality of shots where the subject moved during acquisition.
14 . The non-transitory computer-readable medium of claim 11 , wherein the processor-executable code, when executed by the processor, further cause the processor to:
generate a k-space mask from the k-space data utilizing only the shots not affected by movement of the subject during acquisition, wherein the k-space mask is configured to mask the shots where the subject moved during acquisition; and
apply the k-space mask to the k-space data to generate the dominant pose k-space data.