A Novel Approach to MRI Reconstruction: Complex Neural Networks with Hierarchical Architectures

This paper introduces LCPNN, a novel complex neural network that leverages Laplacian pyramid methodologies to effectively reconstruct magnetic resonance images of superior quality from sparsely sampled k-space data. The study elucidates three pivotal contributions: 1) The implementation of a novel framework that adeptly exploits the advantageous multi-scale attributes of Laplacian pyramid decomposition, 2) the formulation of a sequential multi-scale network structure that integrates intricate convolutions within the aforementioned framework, and 3) the verification of the method's auspicious potential in conserving image edges and intricate textures via experiments conducted on the publicly accessible MICCAI dataset.