Regularization in parallel magnetic resonance imaging

SPIRiT (iterative self‐consistent parallel imaging reconstruction) can be solved efficiently for data acquired on arbitrary k ‐space trajectories, and its sparsity regularized variant L1‐SPIRiT accelerates reconstruction. In this paper, we propose a regularized SPIRiT reconstruction based on steerable pyramid decomposition. The directionally filter banks lead to a better separation of signal and noise compared to a discrete wavelet transform (DWT). In vivo datasets and eight‐channel Shepp‐Logan phantom studies demonstrate efficient reconstructions. We compared our work with five state‐of‐the‐art parallel imaging techniques; our method yields better reconstruction results.