A method and system for reconstructing magnetic resonance (MR) images, the method including the steps of: receiving an under-sampled MR image, the under-sampled MR image being transformed from under-sampled k-spaced data; classifying intensity of each pixel in the under-sampled MR image to one of a plurality of predetermined quantized values of intensity by using a neural network; and generating a reconstructed MR image based on the classified quantized value of the intensity for each pixel in the under-sampled MR image.