HSSRecon partitions data by physical acquisition units and learns a Hankel subspace, achieving competitive reconstruction metrics across six aggregated fastMRI brain conditions
The work proposes HSSRecon, a scan-specific self-supervised reconstruction framework for parallel MRI: it partitions data by physical acquisition units before Hankel lifting and applies multiplicity normalization to repeated Hankel copies in overlapping windows, the network learns a compact complex-valued Hankel subspace operator, and reconstruction is performed over the original k-space variables with a conjugate-gradient solver and hard data consistency; on fastMRI brain data with three contrasts and three sampling masks across six aggregated conditions, it achieves competitive PSNR, SSIM, and NMSE.