Synergized pulsing-imaging network (SPIN)
A synergized pulsing-imaging network is described. A method of optimizing a magnetic resonance imaging (MRI) system includes optimizing, by a synergized pulsing-imaging network (SPIN) circuitry a pulse sequence based, at least in part, on a loss function associated with a reconstruction network. The method further includes optimizing, by the SPIN circuitry, the reconstruction network based, at least in part, on intermediate raw MRI data and based, at least in part, on a ground truth MRI image data. The intermediate raw MRI data is determined based, at least in part on the pulse sequence.
1. A method of optimizing a magnetic resonance imaging (MRI) system, the method comprising:
optimizing, by a synergized pulsing-imaging network (SPIN) circuitry, a pulse sequence based, at least in part, on a loss function associated with a reconstruction network; and
optimizing, by the SPIN circuitry, the reconstruction network based, at least in part, on intermediate raw MRI data and based, at least in part, on a ground truth MRI image data, the intermediate raw MRI data determined based, at least in part, on the pulse sequence.
2. The method of claim 1 , further comprising
determining, by the SPIN circuitry, an intermediate loss function based, at least in part, on the ground truth MRI image data and based, at least in part, on an optimized MRI image data output from the reconstruction network.
3. The method of claim 1 , wherein the optimizing the pulse sequence comprises iteratively adjusting at least one pulse sequence parameter.
4. The method of claim 3 , wherein the at least one pulse sequence parameter is selected from the group comprising a flip angle (α), a first magnetic field gradient (G x ), a second magnetic field gradient (G y ) and a proton density (ρ).
5. The method of claim 3 , wherein the at least one pulse sequence parameter is adjusted based, at least in part, on a loss gradient and based, at least in part, on a learning rate.
6. The method of claim 1 , wherein the optimizing the pulse sequence and the optimizing the reconstruction network are iterative.
7. The method of claim 1 , wherein the optimizing comprises determining whether a current intermediate loss function is within a maximum loss increment of a prior loss function.
8. The method of claim 1 , wherein the pulse sequence is selected from the group comprising a spin-echo (SE) pulse sequence, a gradient-echo (GE) pulse sequence, an echo planar imaging (EPI) pulse sequence and an MR fingerprint (MRF) pulse sequence.
9. The method of claim 1 , wherein the reconstruction network is an artificial neural network (ANN) selected from the group comprising a deep neural network, a convolutional neural network (CNN), a residual encoder-decoder CNN (RED-CNN), a generative adversarial network (GAN) and/or a multilayer perceptron.
10. A magnetic resonance imaging (MRI) system comprising:
a synergized pulsing-imaging network (SPIN) circuitry configured to optimize a pulse sequence based, at least in part, on a loss function associated with a reconstruction network,
the SPIN circuitry further configured to optimize the reconstruction network based, at least in part, on intermediate raw MRI data and based, at least in part, on a ground truth MRI image data, the intermediate raw MRI data determined based, at least in part, on the pulse sequence.
11. The system of claim 10 , wherein the SPIN circuitry is configured to determine an intermediate loss function based, at least in part, on the ground truth MRI image data and based, at least in part, on an optimized MM image data output from the reconstruction network.
12. The system of claim 10 , wherein the optimizing the pulse sequence comprises iteratively adjusting at least one pulse sequence parameter.
13. The system of claim 12 , wherein the pulse sequence parameter is selected from the group comprising a flip angle (α), a first magnetic field gradient (G x ), a second magnetic field gradient (G y ) and a proton density (ρ).
14. The system of claim 12 , wherein the at least one pulse sequence parameter is adjusted based, at least in part, on a loss gradient and based, at least in part, on a learning rate.
15. The system according to claim 10 , wherein the optimizing the pulse sequence and the optimizing the reconstruction network are iterative.
16. The system according to claim 10 , wherein the optimizing comprises determining whether a current intermediate loss function is within a maximum loss increment of a prior loss function.
17. The system according to claim 10 , wherein the pulse sequence is selected from the group comprising a spin-echo (SE) pulse sequence, a gradient-echo (GE) pulse sequence, an echo planar imaging (EPI) pulse sequence and an MR fingerprint (MRF) pulse sequence.
18. The system according to claim 10 , wherein the reconstruction network is an artificial neural network (ANN) selected from the group comprising a deep neural network, a convolutional neural network (CNN), a residual encoder-decoder CNN (RED-CNN), a generative adversarial network (GAN) and/or a multilayer perceptron.
19. A device comprising means for performing the method according to claim 1 .
20. A computer readable storage device having stored thereon instructions that when executed by one or more processors result in the following operations comprising the method according to claim 1 .