System and method for generating a water-fat seperated image
A system for generating a water-fat separated image is provided. The system includes a memory device that stores training results of a neural network, and an MRI controller in electronic communication with the memory device and a magnet assembly. The controller is operative to: acquire MR data from an object via the magnet assembly; generate at least two magnitude images from the acquired MR data; and generate the water-fat separated image by applying the training results to the at least two magnitude images.
1. A system for generating a water-fat separated image, the system comprising:
a memory device that stores training results of a neural network;
the training results generated by supervised learning wherein the output of the neural network is optimized to a water-fat image dataset;
an MRI controller in electronic communication with the memory device and a magnet assembly and operative to:
acquire MR data from an object via the magnet assembly;
generate at least two magnitude images from the acquired MR data; and
generate the water-fat separated image by applying the training results to the at least two magnitude images.
2. The system of claim 1 , wherein the water-fat separated image is a water image.
3. The system of claim 1 , wherein the MR data is acquired from a pulse sequence having two or more echo times.
4. The system of claim 3 , wherein
a first magnitude image of the at least two magnitude images is generated from MR data acquired at a first echo time of the two or more echo times; and
a second magnitude image of the at least two magnitude images is generated from MR data acquired at a second echo time of the two or more echo times.
5. The system of claim 3 , wherein the pulse sequence is a two-point Dixon pulse sequence and the two or more echo times are flexible echo times.
6. The system of claim 1 , wherein the MRI controller applies the training results to the at least two magnitude images via the neural network.
7. The system of claim 6 , wherein the neural network is a convolutional neural network.
8. The system of claim 7 , wherein the neural network is based at least in part on a U-net architecture.
9. A method of training a neural network, the method comprising:
feeding a plurality of magnitude images to the neural network, wherein each magnitude image is generated from a set of MR data;
training a water-fat separation model on the plurality of magnitude images, wherein a loss function based at least in part on a structural dissimilarity metric is used in the training; and
outputting the water-fat separation model as training results.
10. The method of claim 9 , wherein the neural network is a convolutional neural network.
11. The method of claim 10 , wherein the neural network is based at least in part on a U-net architecture.
12. The method of claim 9 further comprising:
transmitting the training results over a network to one or more instantiations of the neural network.
13. A system for generating a water-fat separated image, the system comprising:
a memory device that stores training results of a neural network;
the training results generated by supervised learning wherein the output of the neural network is optimized to a water-fat image dataset;
an MRI controller in electronic communication with the memory device and a magnet assembly and operative to:
acquire MR data from an object via the magnet assembly;
generate a real part image and an imaginary part image from the acquired MR data; and
generate the water-fat separated image by applying the training results to the real part image and the imaginary part image.
14. The system of claim 13 , wherein the MRI controller applies the training results to the real part image and the imaginary part image via the neural network.
15. The system of claim 14 , wherein the neural network is a convolutional neural network.
16. The system of claim 15 , wherein the neural network is based at least in part on a U-net architecture.
17. A method of training a neural network, the method comprising:
feeding a plurality pairs of real part images and imaginary part images to the neural network, wherein each pair of real part image and imaginary part image is generated from a set of MR data;
training a water-fat separation model on the plurality pairs of real part images and imaginary part images, wherein a loss function based at least in part on a structural dissimilarity metric is used in the training; and
outputting the water-fat separation model as training results.
18. The method of claim 17 , wherein the neural network is a convolutional neural network.
19. The method of claim 18 , wherein the neural network is based at least in part on a U-net architecture.
20. The method of claim 17 further comprising:
transmitting the training results over a network to one or more instantiations of the neural network.