Method for processing multi-directional X-ray computed tomography image using artificial neural network and apparatus therefor

1. An image processing method, comprising:

receiving a predetermined number of multi-directional X-ray computed tomography (CT) data; and

reconstructing an image for the multi-directional X-ray CT data using a neural network learned in each of an image domain and a sinogram domain,

wherein the reconstructing of the image includes:

obtaining initial reconstruction images from the multi-directional X-ray CT data using a predetermined analytic algorithm;

denoising the initial reconstruction images using a neural network of the image domain;

generating projection view data by projecting the denoised initial reconstruction images;

obtaining sinogram data denoised from the projection view data using a neural network of the sinogram domain; and

reconstructing the image for the multi-directional X-ray CT data from the denoised sinogram data using the analytic algorithm.

2. The image processing method of claim 1 , wherein the neural network of the image domain denoises the initial reconstruction images using a learning model for mapping between a predetermined artifact-corrupted image and model-based iterative reconstruction (MBIR) instruction in the image domain.

3. The image processing method of claim 1 , wherein the neural network of the sinogram domain provides the sinogram data denoised from the projection view data using a learning model for mapping between sinogram data and projection data in the sinogram domain.

4. The image processing method of claim 1 , wherein the reconstructing of the image includes:

reconstructing the image using an analytic algorithm including a Feldkamp-Davis-Kress (FDK) algorithm and a filtered-backprojection (FBP) algorithm.

5. The image processing method of claim 1 , wherein the neural network includes a convolutional framelet-based neural network.

6. The image processing method of claim 1 , wherein the neural network includes a multi-resolution neural network including a pooling layer and an unpooling layer.

7. The image processing method of claim 6 , wherein the neural network includes a bypass connection from the pooling layer to the unpooling layer.

8. The image processing method of claim 1 , wherein the reconstructing of the image includes:

reconstructing the image for the multi-directional X-ray CT data by using the image reconstructed using a neural network of the image domain and a neural network of the sinogram domain as label data of the neural network of the image domain.

9. An image processing method, comprising:

receiving computed tomography (CT) data; and

reconstructing an image for the CT data using a neural network learned in each of an image domain and a sinogram domain,

wherein the reconstructing of the image includes:

obtaining an initial reconstruction image from the CT data using a predetermined analytic algorithm;

denoising the initial reconstruction image using the neural network of the image domain;

generating projection view data by projecting the denoised initial reconstruction image;

obtaining sinogram data denoised from the projection view data using the neural network of the sinogram domain; and

reconstructing the image for the CT data from the denoised sinogram data using the analytic algorithm.

10. An image processing device, comprising:

a reception unit configured to receive a predetermined number of multi-directional X-ray computed tomography (CT) data; and

a reconstruction unit configured to reconstruct an image for the multi-directional X-ray CT data using a neural network learned in each of an image domain and a sinogram domain,

wherein the reconstruction unit obtains initial reconstruction images from the multi-directional X-ray CT data using a predetermined analytic algorithm, denoises the initial reconstruction images using the neural network of the image domain, generates projection view data by projecting the denoised initial reconstruction images, obtains sinogram data denoised from the projection view data using the neural network of the sinogram domain, and reconstructs the image for the multi-directional X-ray CT data from the denoised sinogram data using the analytic algorithm.

11. The image processing device of claim 10 , wherein the neural network of the image domain denoises the initial reconstruction images using a learning model for mapping between a predetermined artifact-corrupted image and model-based iterative reconstruction (MBIR) instruction in the image domain.

12. The image processing device of claim 10 , wherein the neural network of the sinogram domain provides the sinogram data denoised from the projection view data using a learning model for mapping between sinogram data and projection data in the sinogram domain.

13. The image processing device of claim 10 , wherein the reconstruction unit reconstructs the image using an analytic algorithm including a Feldkamp-Davis-Kress (FDK) algorithm and a filtered-backprojection (FBP) algorithm.

14. The image processing device of claim 10 , wherein the neural network includes a convolutional framelet-based neural network.

15. The image processing device of claim 10 , wherein the neural network includes a multi-resolution neural network including a pooling layer and an unpooling layer.

16. The image processing device of claim 15 , wherein the neural network includes a bypass connection from the pooling layer to the unpooling layer.

17. The image processing device of claim 10 , wherein the reconstruction unit reconstructs the image for the multi-directional X-ray CT data by using the image reconstructed using a neural network of the image domain and a neural network of the sinogram domain as label data of the neural network of the image domain.