Magnetic resonance imaging method and device
Methods and devices for magnetic resonance imaging are provided. In one aspect, a method includes: obtaining undersampled k-space data as first partial k-space data by scanning a subject in an accelerated scanning manner, generating a first image by performing image reconstruction for the first partial k-space data according to a trained deep neural network and an explicit analytic solution imaging algorithm, obtaining mapped data of complete k-space by mapping the first image to k-space, extracting second partial k-space data from the mapped data of complete k-space, the second partial k-space data being distributed in the k-space at a same position as the first partial k-space data in the k-space, obtaining a residual image by performing image reconstruction according to the first partial k-space data and the second partial k-space data, and finally generating a magnetic resonance image of the subject by adding the first image with the residual image.
1. A magnetic resonance imaging method comprising:
obtaining undersampled k-space data as first partial k-space data by scanning a subject in an accelerated scanning manner;
generating a first image by performing image reconstruction for the first partial k-space data according to a trained first deep neural network and an explicit analytic solution imaging algorithm;
obtaining mapped data of complete k-space by mapping the first image to k-space;
extracting second partial k-space data from the mapped data of complete k-space, wherein the second partial k-space data is distributed in the k-space at a same position as the first partial k-space data in the k-space;
obtaining a residual image by performing image reconstruction according to the first partial k-space data and the second partial k-space data; and
generating a magnetic resonance image of the subject by adding the first image with the residual image.
2. The method of claim 1 , wherein generating the first image comprises:
obtaining a second image by performing partial reconstruction for the first partial k-space data with the explicit analytic solution imaging algorithm; and
generating the first image by performing complete reconstruction for the second image with the first deep neural network.
3. The method of claim 2 , wherein the accelerated scanning manner comprises one of random undersampling and variable-density undersampling, and
wherein the explicit analytic solution imaging algorithm comprises a GeneRalized Autocalibrating Partially Parallel Acquisition (GRAPPA) algorithm.
4. The method of claim 2 , wherein the accelerated scanning manner comprises uniform undersampling, and
wherein the explicit analytic solution imaging algorithm comprises a SENSitivity Encoding (SENSE) algorithm.
5. The method of claim 2 , wherein obtaining a second image by performing partial reconstruction for the first partial k-space data with the explicit analytic solution imaging algorithm comprises:
generating fitting data based on the first partial k-space data by using a parallel imaging algorithm;
obtaining a particular partial k-space data based on the fitting data and the first partial k-space data by using a half-Fourier imaging algorithm; and
generating the second image by performing the partial reconstruction based on the particular partial k-space data.
6. The method of claim 1 , wherein generating the first image comprises:
obtaining a second image by performing partial reconstruction for the first partial k-space data with the first deep neural network; and
generating the first image by performing complete reconstruction for the second image with the explicit analytic solution imaging algorithm.
7. The method of claim 1 , wherein obtaining the residual image by performing reconstruction according to the first partial k-space data and the second partial k-space data comprises:
obtaining residual k-space data by subtracting the first partial k-space data from the second partial k-space data; and
generating the residual image by performing sparse constraint reconstruction for the residual k-space data.
8. The method of claim 7 , wherein generating the residual image comprises:
performing the sparse constraint reconstruction for the residual k-space data in combination with a parallel imaging algorithm, with an image support domain as a regularization term.
9. The method of claim 1 , wherein obtaining mapped data of complete k-space by mapping the first image to k-space comprises:
obtaining the mapped data of complete k-space by mapping the first image to the k-space based on Fourier transform.
10. The method of claim 9 , wherein mapping the first image to the k-space based on Fourier transform comprises:
multiplying the first image by a coil sensitivity map of a particular channel of a multi-channel array coil that is used for scanning the subject.
11. The method of claim 1 , further comprising:
obtaining a final magnetic resonance image of the subject by denoising the magnetic resonance image.
12. The method of claim 11 , wherein denoising the magnetic resonance image comprises:
denoising the magnetic resonance image with a trained second deep neural network.
13. The method of claim 12 , wherein the trained second deep neural network is different from the trained first deep neural network.
14. A magnetic resonance imaging device, comprising:
at least one processor; and
a memory configured to store machine readable instructions corresponding to a control logic for magnetic resonance imaging;
wherein the processor is configured to read the machine readable instructions in the memory and execute the instructions to perform operations comprising:
obtaining undersampled k-space data as a first partial k-space data by scanning a subject in an accelerated scanning manner;
generating a first image by performing image reconstruction for the first partial k-space data according to a trained first deep neural network and an explicit analytic solution imaging algorithm;
obtaining mapped data of complete k-space by mapping the first image to k-space;
extracting second partial k-space data from the mapped data of complete k-space, wherein the second partial k-space data is distributed in the k-space at a same position as the first partial k-space data in the k-space;
obtaining a residual image by performing image reconstruction according to the first partial k-space data and the second partial k-space data; and
generating a magnetic resonance image of the subject by adding the first image with the residual image.
15. The device of claim 14 , wherein obtaining mapped data of complete k-space by mapping the first image to the k-space comprises:
obtaining the mapped data of complete k-space by mapping the first image to the k-space based on Fourier transform.
16. The device of claim 14 , wherein the operations further comprises:
obtaining a final magnetic resonance image of the subject by denoising the magnetic resonance image.
17. The device of claim 16 , wherein denoising the magnetic resonance image comprises:
denoising the magnetic resonance image with a trained second deep neural network that is different from the trained first deep neural network.
18. The device of claim 14 , wherein obtaining the residual image by performing reconstruction according to the first partial k-space data and the second partial k-space data comprises:
obtaining residual k-space data by performing subtraction for the second partial k-space data and the first partial k-space data; and
generating the residual image by performing sparse constraint reconstruction for the residual k-space data.
19. The device of claim 18 , wherein generating the residual image by performing sparse constraint reconstruction for the residual k-space data comprises:
with an image support domain as a regularization term, generating the residual image by performing the sparse constraint reconstruction for the residual k-space data in combination with a parallel imaging algorithm.
20. The device of claim 14 , wherein generating the first image by performing image reconstruction for the first partial k-space data according to the trained first deep neural network and the explicit analytic solution imaging algorithm comprises:
obtaining a second image by performing partial reconstruction for the first partial k-space data with the explicit analytic solution imaging algorithm; and
generating the first image by performing complete reconstruction for the second image with the first deep neural network.