Method for reconstructing QSM MAP using dipole compensation
View Patent ↗A method of generating a quantitative susceptibility mapping (QSM) map capable of reconstructing data having various resolutions using a given network trained at a specific resolution includes preparing a local field map having input resolution resol input , generating a plurality of sub-images having training resolution resol train by resampling the local field map, inputting each of the sub-images into QSMnet_resol train , which is an inference network trained with the training resolution, acquiring a plurality of QSM sub-images having the training resolution resol train from the inference network, generating one QSM assembled map by assembling the plurality of QSM sub-images, and generating a QSM map (corrected QSM map) from the QSM assembled map by applying a dipole compensation method.
1. A method of generating a QSM map comprising:
acquiring, by an MRI system, an image having a second resolution by scanning an object using an MRI scanner;
generating, by the MRI system, a plurality of sub-images having a first resolution from the image;
generating, by the MRI system, a plurality of QSM sub-maps having the first resolution from the plurality of sub-images using an inference network;
generating, by the MRI system, one QSM assembled map having the second resolution by assembling the plurality of QSM sub-maps with each other, and
generating, by the MRI system, a QSM map having the second resolution by calibrating the one QSM assembled map,
wherein,
the calibration is performed based on a difference between a second dipole kernel represented by the second resolution and a third dipole kernel represented by the second resolution, and
the third dipole kernel is obtained by performing sampling conversion on a first dipole kernel represented by the first resolution so that the first dipole kernel is represented by the second resolution.
2. The method of claim 1 , wherein
the generating of the QSM map comprises
generating, by the MRI system, a first QSM assembled map frequency signal by Fourier transforming the generated one QSM assembled map,
generating, by the MRI system, a first QSM map frequency signal by calibrating the generated first QSM assembled map frequency signal based on a difference between the second dipole kernel and the third dipole kernel, and
generating, by the MRI system, a QSM map having the second resolution by Fourier transforming the generated first QSM map frequency signal.
3. The method of claim 1 , wherein
the second resolution is larger than the first resolution,
a first frequency signal, which is a frequency domain signal of the first dipole kernel, is defined in a section where kx is [−b1, b1],
both a second frequency signal, which is a frequency domain signal of the second dipole kernel, and a third frequency signal, which is a frequency domain signal of the third dipole kernel, are defined in a section where kx is [−b2, b2] (b2>b1),
the second frequency signal and the third frequency signal are equal to each other in a section where kx is [−b1, b1], and the second frequency signal and the third frequency signal are different from each other in a section where kx is [−b2, −b1] and a section where kx is [b1, b2], and
the calibration is performed in the section where kx is [−b2, −b1] and the section where kx is [b1, b2].
4. The method of claim 3 , wherein
a value of the first QSM map frequency signal at kx=k is a value obtained by multiplying the value of the first QSM assembled map frequency signal at kx=k by a predetermined value,
the predetermined value is a value obtained by dividing a value of the third frequency signal at kx=k by a value of the second frequency signal at kx=k, and
the k is a value belonging to the section where kx is [−b2, −b1] and the section where kx is [b1, b2].
5. The method of claim 1 , wherein
the inference network is trained by supervised learning using a plurality of input images for training having the first resolution and using a plurality of QSM maps generated from the plurality of input images for training as labels.
6. The method of claim 1 , wherein
all of the plurality of sub-images have the same horizontal and vertical size,
all of the plurality of QSM sub-maps have the same horizontal and vertical size, and
the second resolution is larger than the first resolution.
7. The method of claim 6 , wherein
a plurality of pixels included in a first region of the QSM assembled map include a plurality of pixels located at a first position corresponding to each other in the plurality of QSM sub-maps.
8. The method of claim 1 , wherein
the plurality of sub-images are different images obtained by performing downsampling on the image.
9. A method of generating a QSM map comprising:
acquiring, by an MRI system, an image having first resolution by scanning an object using an MRI scanner;
generating, by the MRI system, one upsampling image having second resolution larger than the first resolution from the image;
generating, by the MRI system, one QSM upsampling map having the second resolution from the upsampling image using an inference network; and
generating, by the MRI system, one QSM map having the first resolution by calibrating the one QSM upsampling map having the second resolution.
10. An MRI system comprising:
an MRI scanner; and
a computing device, wherein
the computing device is configured to execute
acquiring an image having second resolution by scanning an object using the MRI scanner,
generating a plurality of sub-images having first resolution from the image,
generating a plurality of QSM sub-maps having the first resolution from the plurality of sub-images using an inference network,
generating one QSM assembled map having the second resolution by assembling the plurality of QSM sub-maps with each other, and
generating a OSM map having the second resolution by calibrating the one QSM assembled map,
the calibration is performed based on a difference between a second dipole kernel represented by the second resolution and a third dipole kernel represented by the second resolution, and
the third dipole kernel is obtained by performing sampling conversion on a first dipole kernel represented by the first resolution so that the first dipole kernel is represented by the second resolution.
11. The MRI system of claim 10 , wherein
the generating of the QSM map comprises
generating, by the MRI system, a first QSM assembled map frequency signal by Fourier transforming the generated one QSM assembled map,
generating, by the MRI system, a first QSM map frequency signal by calibrating the generated first QSM assembled map frequency signal based on a difference between the second dipole kernel and the third dipole kernel, and
generating, by the MRI system, a QSM map having the second resolution by Fourier transforming the generated first QSM map frequency signal.
12. The MRI system of claim 10 , wherein
the inference network is trained by supervised learning using a plurality of input images for training having the first resolution and using a plurality of QSM maps generated from the plurality of input images for training as labels.
13. A method of generating a QSM map comprising:
acquiring, by a computing device, an image having second resolution by scanning an object using an MRI scanner;
generating, by the computing device, a plurality of sub-images having first resolution from the image;
generating, by the computing device, a plurality of QSM sub-maps having the first resolution from the plurality of sub-images using an inference network;
generating, by the computing device, one QSM assembled map having the second resolution by assembling the plurality of QSM sub-maps; and
generating, by the computing device, a QSM map having the second resolution by calibrating the one QSM assembled map, wherein
the calibration is performed based on a difference between a second dipole kernel represented by the second resolution and a third dipole kernel represented by the second resolution, and
the third dipole kernel is obtained by performing sampling conversion on a first dipole kernel represented by the first resolution so that the first dipole kernel is represented by the second resolution.