Apparatus and method for reconstructing magnetic resonance image using learning, and under-sampling apparatus method and recording medium thereof

An MR image reconstruction apparatus using learning, comprising:

• a decomposition portion that generates n×k1 low-resolution input images by applying k1 first filters n times to a high-resolution input image;
• a learning portion that generates n×k2 low-resolution output images by applying a corresponding neural network according to the n×k1 low-resolution input image;
• and
• a reconstruction portion that generates a high-resolution output image by applying k2 second dual filters n times to the n×k2 low-resolution output image,
• wherein the high-resolution input image is generated by converting a k-space image where full-sampling is performed on a first region that corresponds to a center of the k-space image and under-sampling is performed on a second region in the k-space image, excluding the first region, and
• the k2 second dual filters are dual filters with respect to k2 second filters that generate a reference high-resolution label image for learning of the neural network as n×k2 low-resolution label images.

A MR image reconstruction method using learning, comprising:

• under-sampling a first k-space image by full-sampling a first region that corresponds to a center of the first k-space image and under-sampling a second region, excluding the first region, in the first k-space image;
• outputting an MR image based on the first k-space image through a pre-learned neural network;
• generating a third k-space image by correcting a second k-space image that is based on the output MR image, based on the first k-space image with respect to a region corresponding to the first k-space image in the second k-space image;
• and
• acquiring an MR image based on the third k-space image.

An MR image reconstruction method using learning, comprising:

• generating a high-resolution input image by converting a k-space image where full-sampling is performed on a first region that corresponds to a center of the k-space image and under-sampling is performed on a second region in the k-space image, excluding the first region;
• decomposing for generating n×k1 low-resolution input images by applying k1 first filters on the high-resolution input image;
• generating n×k2 low-resolution output images by applying a corresponding neural network according to the n×k1 low-resolution input images;
• and
• generating a high-resolution output image by applying nk2 second dual filters to the n×k2 low-resolution output images,
• wherein the k2 second dual filters are dual filters with respect to k2 second filters that generate a reference high-resolution label image for learning of the neural network as n×k2 low-resolution label images.