Systems and methods for magnetic resonance imaging standardization using deep learning

1. A computer-implemented method for transforming magnetic resonance (MR) imaging across multiple vendors comprising:

(a) during a training phase, obtaining a training dataset, wherein the training dataset comprises a paired dataset and an un-paired dataset, wherein the training dataset comprises image data acquired using two or more MR imaging devices provided by two or more different vendors, and wherein the image data have different image styles of MR imaging;

(b) training a deep network model to learn a function that transforms images of different image styles to a target image style, wherein the training comprises using a combination of supervised learning and unsupervised learning, and wherein the supervised learning comprises using the paired data set of the training dataset to train the deep network model and the unsupervised learning comprises using the un-paired dataset to train the deep network model;

(c) after the deep network model is trained, using the trained deep network model to generate a synthesized MR image by processing an input MR image, wherein the synthesized MR image has the target image style and the input MR image has an image style different from the target image style; and

(d) displaying the synthesized MR image with the target image style on a display device.

2. The computer-implemented method of claim 1 , wherein the image data acquired using the two or more MR imaging devices save different image styles.

3. The computer-implemented method of claim 1 , wherein the plurality of different image styles are different in at least one of contrast, resolution and image distortion.

4. The computer-implemented method of claim 1 , wherein the target image style is pre-determined or selected based on a set of pre-determined rules.

5. The computer-implemented method of claim 4 , wherein the target image style comprises one or more characteristics including contrast, resolution or image distortion.

6. The computer-implemented method of claim 1 , wherein the target image style is consistent with an image style corresponding to a given MR imaging device, wherein the given MR imaging device is different from an MR imaging device used for acquiring the input MR image.

7. The computer-implemented method of claim 1 , wherein the paired dataset comprises reference image data acquired using a first MR imaging device and an original image data acquired using a second MR imaging device that is different from the first MR imaging device.

8. The computer-implemented method of claim 1 , wherein the combination of supervised learning and unsupervised learning comprises training the deep network model using the supervised learning and further enhancing the deep network model using the unsupervised learning.

9. The computer-implemented method of claim 1 , wherein the unsupervised learning comprises using Cycle Generative Adversarial Network.

10. The computer-implemented method of claim 1 , wherein the deep learning model comprises U-net neural network structures.

11. The computer-implemented method of claim 1 , wherein the target image style is determined by a user via a graphical user interface.

12. A non-transitory computer-readable storage medium including instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising:

(a) during a training phase, obtaining a training dataset, wherein the training dataset comprises a paired dataset and an un-paired dataset, wherein the training dataset comprises image data acquired using two or more MR imaging devices provided by two or more different vendors, and wherein the image data comprises different image styles of MR imaging;

(b) training a deep network model to learn a function that transforms images of different image styles to a target image style, wherein the training comprises using a combination of supervised learning and unsupervised learning, and wherein the supervised learning comprises using the paired data set of the training dataset and the unsupervised learning comprises using the un-paired dataset to train the deep network model;

(c) after the deep network model is trained, using the trained deep network model to generate a synthesized MR image by processing an input MR image, wherein the synthesized MR image has the target image style and the input MR image has an image style different from the target image style; and

(d) displaying the synthesized MR image with the target image style on a display device.

13. The non-transitory computer-readable storage medium of claim 12 , wherein the image data acquired using the two or more MR imaging devices have different image styles.

14. The non-transitory computer-readable storage medium of claim 12 , wherein the plurality of different image styles are different in at least one of contrast, resolution and image distortion.

15. The non-transitory computer-readable storage medium of claim 12 , wherein the target image style is pre-determined or selected based on a set of pre-determined rules.

16. The non-transitory computer-readable storage medium of claim 15 , wherein the target image style comprises one or more characteristics including contrast, resolution or image distortion.

17. The non-transitory computer-readable storage medium of claim 12 , wherein the target image style corresponds to an image style of a given MR imaging device, wherein the given MR imaging device is different from a MR imaging device used for acquiring the input MR image.

18. The non-transitory computer-readable storage medium of claim 12 , wherein the paired dataset comprises reference image data acquired using a first MR imaging device and an original image data acquired using a second MR imaging device that is different from the first MR imaging device.

19. The non-transitory computer-readable storage medium of claim 12 , wherein the combination of supervised learning and unsupervised learning comprises training the deep network model using the supervised learning and further enhancing the deep network model using the unsupervised learning.

20. The non-transitory computer-readable storage medium of claim 12 , wherein the unsupervised learning comprises using Cycle Generative Adversarial Network.