Method for processing sparse-view computed tomography image using neural network and apparatus therefor

1. An image processing method, comprising:

receiving a sparse-view computed tomography (CT) data; and

reconstructing an image for the sparse-view CT data using a neural network of a learning model satisfying a predetermined frame condition,

wherein the neural network comprises:

a multi-resolution neural network including pooling and unpooling layers; and

a structured tight frame neural network by decomposing a structured dual frame neural network and the multi-resolution neural network into a low-frequency domain and a high-frequency domain using wavelets by expressing a mathematical expression of the multi-resolution neural network as a dual frame {tilde over (Φ)},

wherein the predetermined frame condition is represented by the equation

{tilde over (Φ)}Φ T =I.

2. The image processing method of claim 1 , wherein the reconstructing of the image comprises:

reconstructing the image for the sparse-view CT data using the neural network of the learning model which satisfies the frame condition and is learned by residual learning.

3. The image processing method of claim 1 , wherein the neural network comprises:

a neural network which generates the learning model satisfying the frame condition through a mathematical analysis based on convolutional framelets and is learned by the learning model.

4. The image processing method of claim 1 , wherein the neural network comprises:

a by-pass connection from the pooling layer to the unpooling layer.

5. An image processing method, comprising:

receiving a sparse-view CT data; and

reconstructing an image for the sparse-view CT data using a neural network for a learning model which satisfies a predetermined frame condition and is based on convolutional framelets,

wherein the neural network comprises:

a multi-resolution neural network including pooling and unpooling layers; and

a structured tight frame neural network by decomposing a structured dual frame neural network and the multi-resolution neural network into a low-frequency domain and a high-frequency domain using wavelets by expressing a mathematical expression of the multi-resolution neural network as a dual frame {tilde over (Φ)},

wherein the predetermined frame condition is represented by the equation

{tilde over (Φ)}Φ T =I.

6. An image processing device, comprising:

a reception unit configured to receive a sparse-view CT data; and

a reconstruction unit configured to reconstruct an image for the sparse-view CT data using a neural network of a learning model satisfying a predetermined frame condition,

wherein the neural network comprises:

a multi-resolution neural network including pooling and unpooling layers; and

a structured tight frame neural network by decomposing a structured dual frame neural network and the multi-resolution neural network into a low-frequency domain and a high-frequency domain using wavelets by expressing a mathematical expression of the multi-resolution neural network as a dual frame {tilde over (Φ)},

wherein the predetermined frame condition is represented by the equation

{tilde over (Φ)}Φ T =I.

7. The image processing device of claim 6 , wherein the reconstruction unit is configured to:

reconstruct the image for the sparse-view CT data using the neural network of the learning model which satisfies the frame condition and is learned by residual learning.

8. The image processing device of claim 6 , wherein the neural network comprises:

a neural network which generates the learning model satisfying the frame condition through a mathematical analysis based on convolutional framelets and is learned by the learning model.

9. The image processing device of claim 6 , wherein the neural network comprises:

a by-pass connection from the pooling layer to the unpooling layer.