SYSTEM AND METHOD OF GENERATING MAGNETIC RESONANCE IMAGE USING DEEP DEEP ARTIFICIAL NEURAL NETWORK

摘要

The present invention relates to a magnetic resonance image generation system and method for reconstructing high-quality magnetic resonance images having high spatial and temporal resolution from data acquired at a sampling rate lower than the Nyquist rate using a deep artificial neural network. will be. A magnetic resonance image generation method according to the present invention comprises the steps of: (a) generating an initial image for learning from compressed sensing data for learning generated from full data for learning; (b) generating an error image for learning that is a difference between the reference image for learning generated from the full data for learning and the initial image for learning; (c) training a deep artificial neural network with a first t-y plane image for training obtained by slicing the initial image for training and a second t-y plane image for training obtained by slicing the error image for training; (d) generating an initial image of the object from the object compression sensing data obtained from the object; (e) generating a second object t-y plane image by providing a first object t-y plane image obtained by slicing the initial object image to the deep artificial neural network learned in step (c); and (f) generating an estimated object image from the object error image obtained from the second object t-y plane image and the initial object image.