K-space refinement in deep learning MR reconstruction via regularizing scan specific SPIRiT-based self consistency

Deep Learning (DL) based reconstruction using unrolled neural networks has shown great potential in accelerating magnetic resonance imaging (MRI). However, one of the major drawbacks is the loss of high-frequency details and textures in the output. In this paper, we propose a novel refinement method based on SPIRiT (Iterative Self-consistent Parallel Imaging Reconstruction from Arbitrary k-Space) formulation to reduce the k-space errors and enable reconstruction of improved high-frequency image details and textures. The proposed scheme constrains the DL output to satisfy the neighborhood relationship in the frequency space (k-space) which can be easily calibrated in the auto-calibration (ACS) lines, and corrects the underestimation in the peripheral region of the k-space as well as reduce structured k-space errors. We show that our method enables the reconstruction of sharper images with significantly improved high-frequency components measured by HFEN and GMSD while maintaining overall error in the image measured by PSNR and SSIM.