Robust subspace-based parallel imaging (SPAN): Bridging sensitivity maps and convolution kernels

Motivation: The reconstruction algorithms that rely on explicit sensitivity maps, like SENSE, are not robust, and prone to producing artifacts. Goal(s): Our goal was to devise a novel, robust reconstruction method that is less susceptible to the inaccuracy of sensitivity maps. Approach: The proposed method employs convolution kernels, derived from the singular value decomposition of block-wise Hankel matrix constructed from the k-space of sensitivity maps, to reconstruct under-sampled data. Results: The proposed approach demonstrated superior and more robust performance compared to the SENSE method. Impact: Transforming sensitivity maps into convolution kernels for parallel imaging can enhance image quality and robustness, and provide a novel perspective on leveraging sensitivity maps for researchers.