Systems and methods for improving magnetic resonance imaging using deep learning
A computer-implemented method is provided for improving image quality with shortened acquisition time. The method comprises: determining an accelerated image acquisition scheme for imaging a subject using a medical imaging apparatus; acquiring a medical image of the subject according to the accelerated image acquisition scheme using the medical imaging apparatus; applying a deep network model to the medical image to improve the quality of the medical image; and outputting an improved quality image of the subject, for analysis by a physician.
1. A computer-implemented method for improving image quality with shortened acquisition time, the method comprising:
(a) acquiring a first medical image having a first image quality, wherein the first medical image is a magnetic resonance (MR) medical image of a subject;
(b) applying a deep network model to the first medical image to generate a second medical image, wherein the deep learning model is a deep residual learning model, wherein the second medical image has a second image quality, wherein the second image quality is higher than the first image quality, and wherein the second image quality has greater signal-to-noise ratio (SNR), higher resolution, or less aliasing compared with the first image quality; and
wherein the deep network model is trained by adaptively tuning one or more model parameters to approximate a reference image.
2. The computer-implemented method of claim 1 , wherein the deep learning model is trained using training datasets comprising at least a pair of low quality image and the reference image.
3. The computer-implemented method of claim 2 , wherein the low quality image is generated by applying one or more filters or adding synthetic noise to the reference image to create noise or undersampling artifacts.
4. The computer-implemented method of claim 2 , wherein the low quality image is acquired using an imaging apparatus with shortened acquisition time, undersampled k-space or reduced number of repetitions.
5. The computer-implemented method of claim 2 , wherein signal-to-noise ratio (SNR) or resolution of the reference image is higher than that of the low quality image.
6. The computer-implemented method of claim 2 , wherein the training datasets comprise patches of the pair of low quality image and the reference image.
7. The computer-implemented method of claim 6 , wherein the patches are selected based on one or more metrics quantifying an image similarity.
8. The computer-implemented method of claim 1 , wherein the first medical image is acquired with a shortened acquisition time.
9. The computer-implemented method of claim 1 , wherein the first medical image comprises undersampled k-space image or image acquired using reduced number of repetitions.
10. 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) acquiring a first medical image having a first image quality, wherein the first medical image is a magnetic resonance (MR) medical image of a subject;
(b) applying a deep network model to the first medical image to generate a second medical image, wherein the deep learning model is a deep residual learning model, wherein the second medical image has a second image quality, wherein the second image quality is higher than the first image quality, and wherein the second image quality has greater signal-to-noise ratio (SNR), higher resolution, or less aliasing compared with the first image quality;
and
wherein the deep network model is trained by adaptively tuning one or more model parameters to approximate a reference image.
11. The non-transitory computer-readable storage medium of claim 10 , wherein the deep learning model is trained using training datasets comprising at least a pair of low quality image and the reference image.
12. The non-transitory computer-readable storage medium of claim 11 , wherein the low quality image is generated by applying one or more filters or adding synthetic noise to the reference image to create noise or undersampling artifacts.
13. The non-transitory computer-readable storage medium of claim 11 , wherein the low quality image is acquired using an imaging apparatus with shortened acquisition time, undersampled k-space or reduced number of repetitions.
14. The non-transitory computer-readable storage medium of claim 11 , wherein signal-to-noise ratio (SNR) or resolution of the reference image is higher than that of the low quality image.
15. The non-transitory computer-readable storage medium of claim 11 , wherein the training datasets comprise patches of the pair of low quality image and the reference image.
16. The non-transitory computer-readable storage medium of claim 11 , wherein the first medical image comprises undersampled k-space image, image acquired using reduced number of repetitions, or image acquired with shortened acquisition time.