Free breathing dynamic contrast enhanced (DCE) liver MR imaging
A method for reconstructing dynamic contrast-enhanced (DCE) MR images includes receiving a plurality of continuous free-breathing DCE images, the plurality of images obtained with a contrast, sorting the images by identifying a respiratory phase associated with each of the continuous free-breathing DCE images, reconstructing the plurality continuous free-breathing DCE images into a 4D respiratory motion-resolved image, obtaining 3D deformable motion vector fields (MVFs), and utilizing a deep learning based motion transformation integrated forward-Fourier (DL-MOTIF) model and the 3D deformable MVFs to reconstruct the DCE MR images.
1. A method for reconstructing dynamic contrast-enhanced (DCE) MR images comprising:
receiving a plurality of continuous free-breathing DCE images, the plurality of images obtained with a contrast;
sorting the images by identifying a respiratory phase associated with each of the continuous free-breathing DCE images;
reconstructing the plurality continuous free-breathing DCE images into a 4 D respiratory motion-resolved image;
obtaining 3D deformable motion vector fields (MVFs); and
utilizing a deep learning based motion transformation integrated forward-Fourier (DL-MOTIF) model and the 3D deformable MVFs to reconstruct the DCE MR images.
2. The method of claim 1 , further comprising repeating the method for a plurality of contrasts.
3. The method of claim 1 , wherein the plurality of continuous free-breathing DCE images are obtained using a self-navigated radial golden-angle stack-of-stars sequence.
4. The method of claim 1 , wherein a temporal resolution of the DCE MR images is 10 seconds.
5. The method of claim 1 , wherein the plurality of free-breathing DCE images are sorted into one of five respiratory phases.
6. The method of claim 1 , wherein the DL-MOTIF utilizes a deep learning residual neural network.
7. The method of claim 1 , wherein the 3D MVFs are obtained by generating Phase2Phase (P2P) reconstructed 4 D respiratory motion resolved images from the plurality of images and performing non-linear deformable registration on the 4 D motion resolved images.
8. The method of claim 1 , wherein the plurality of images includes undersampled k-space data.
9. The method of claim 8 , wherein a Nyquist rate of the k-space data is less than 20%.
10. A system for reconstructing dynamic contrast-enhanced (DCE) MR images, the system including at least one processor in communication with at least one memory, the at least one processor configured to:
receive a plurality of continuous free-breathing DCE images, the plurality of images obtained with a contrast;
sort the images by identifying a respiratory phase associated with each of the continuous free-breathing DCE images;
reconstruct the plurality continuous free-breathing DCE images into a 4 D respiratory motion-resolved image;
obtain 3D deformable motion vector fields (MVFs); and
utilize a deep learning based motion transformation integrated forward-Fourier (DL-MOTIF) and the 3D deformable MVFs to reconstruct the DCE MR images.
11. The system of claim 10 , further comprising repeating the method for a plurality of contrasts.
12. The system of claim 10 , wherein the plurality of continuous free-breathing DCE images are obtained using a self-navigated radial golden-angle stack-of-stars sequence.
13. The system of claim 10 , wherein a temporal resolution of the DCE MR images is 10 seconds.
14. The system of claim 10 , wherein the plurality of free-breathing DCE images are sorted into one of five respiratory phases.
15. The system of claim 10 , wherein the DL-MOTIF utilizes a deep learning residual neural network.
16. The system of claim 10 , wherein the 3D MVFs are obtained by generating Phase2Phase (P2P) reconstructed 4 D respiratory motion resolved images from the plurality of images and performing non-linear deformable registration on the 4 D motion resolved images.
17. The system of claim 10 , wherein the plurality of images includes undersampled k-space data.
18. The system of claim 17 , wherein a Nyquist rate of the k-space data is less than 20%.
19. A method for reconstructing dynamic contrast-enhanced (DCE) MR images comprising:
training a deep learning based motion transformation integrated forward-Fourier (DL-MOTIF) model using severely undersampled k-space data;
receiving a plurality of continuous free-breathing DCE images; and
reconstructing the DCE MR images with the DL-MOTIF model.
20. The method of claim 19 , wherein a Nyquist rate of the k-space data is less than 20%.