Systems and methods for reconstructing a medical image using meta learning

1. A computer-implemented method for reconstructing an image, the method comprising:

receiving k-space data;

receiving a transform operator corresponding to the k-space data;

determining a distribution representing information associated with one or more previous iteration images;

generating a next iteration image by an image reconstruction model to reduce an objective function, the objective function corresponding to a data consistency metric and a regularization metric;

evaluating whether the next iteration image is satisfactory; and

if the next iteration image is satisfactory, outputting the next iteration image as an output image;

wherein the data consistency metric corresponds to a first previous iteration image, the k-space data, and the transform operator; and

wherein the regularization metric corresponds to the distribution.

2. The computer-implemented method of claim 1 ,

wherein the image reconstruction model includes an input gate and a forget gate; and

wherein the generating the next iteration image includes updating the image reconstruction model by at least:

updating the input gate based at least in part on the first previous iteration image, an input gate weight, and an input gate bias; and

updating the forget gate based at least in part on the first previous iteration image, a forget gate weight, and a forget gate bias.

3. The computer-implemented method of claim 2 , wherein the input gate is configured to control which one or more new values associated with the one or more previous iteration images is incorporated into the next iteration image.

4. The computer-implemented method of claim 2 , wherein the forget gate is configured to control which one or more previous values associated with the first previous iteration image is incorporated into the next iteration image.

5. The computer-implemented method of claim 2 , wherein the input gate corresponds to a gradient direction and an iterative step size.

6. The computer-implemented method of claim 2 , wherein the forget gate corresponds to an initialization value.

7. The computer-implemented method of claim 2 , wherein the updating the input gate includes updating at least one of the input gate weight and the input gate bias.

8. The computer-implemented method of claim 2 , wherein the updating the forget gate includes updating at least one of the forget gate weight and the forget gate bias.

9. The computer-implemented method of claim 1 , wherein the generating a next iteration image to reduce an objective function includes generating the next iteration image to minimize the objective function.

10. The computer-implemented method of claim 1 , wherein the distribution includes a distribution of sparsity representing the one or more previous iteration images.

11. The computer-implemented method of claim 1 , wherein the distribution includes a distribution of smoothness representing the one or more previous iteration images.

12. The computer-implemented method of claim 1 , wherein the distribution is a normalized distribution having a mean of zero and a variance of unity.

13. The computer-implemented method of claim 1 , wherein the determining a distribution representing information associated with the one or more previous iteration images includes normalizing the information by transforming one or more information values.

14. The computer-implemented method of claim 1 , wherein the distribution is represented by a convolutional neural network configured to represent information associated with the one or more previous iteration images within one or more data consistency layers and one or more skip connections for residual learning.

15. The computer-implemented method of claim 14 , wherein the image reconstruction model uses a long short-term memory model configured to update one or more parameters of the convolutional neural network between iterations.

16. The computer-implemented method of claim 15 , further comprising training the long short-term memory model interactively with training data.

17. The computer-implemented method of claim 1 , wherein evaluating the next iteration image includes evaluating the next iteration image using at least one of SSIM, PSNR, NMSE, and visual inspection.

18. The computer-implemented method of claim 1 , wherein the transform operator includes one of a Fourier transform matrix, an undersampling matrix, and a coil sensitivity map matrix corresponding to a parallel imaging setup.

19. A system for reconstructing an image, the system comprising:

a k-space data receiving module configured to receive k-space data;

a transform operator receiving module configured to receive a transform operator corresponding to the k-space data;

a distribution determining module implemented by the or more processors and configured to determine a distribution representing information associated with one or more previous iteration images;

an image generating module implemented by the one or more processors and configured to generate a next iteration image by an image reconstruction model to reduce an objective function, the objective function corresponding to a data consistency metric and a regularization metric;

an evaluation module implemented by the one or more processors and configured to evaluate whether the next iteration image is satisfactory; and

an outputting module configured to, if the next iteration image is satisfactory, output the next iteration image as an output image;

wherein the data consistency metric corresponds to a first previous iteration image, the k-space data, and the transform operator; and

wherein the regularization metric corresponds to the distribution.

20. A non-transitory computer-readable medium with instructions stored thereon, that when executed by a processor, causes the processor to perform the processes including:

receiving k-space data;

receiving a transform operator corresponding to the k-space data;

determining a distribution representing information associated with one or more previous iteration images;

generating a next iteration image by an image reconstruction model to reduce an objective function, the objective function corresponding to a data consistency metric and a regularization metric;

evaluating whether the next iteration image is satisfactory; and

if the next iteration image is satisfactory, outputting the next iteration image as an output image;

wherein the data consistency metric corresponds to a first previous iteration image, the k-space data, and the transform operator; and

wherein the regularization metric corresponds to the distribution.