Multi-coil undersampled MRI reconstruction combining Wave-CAIPI sparse sampling and statistical distribution prior

Objective In12-fold accelerated multi-coil brain MRI, joint reconstruction of multi-channel data faces a severely underdetermined problem due to high-density k-space undersampling, leading to significant aliasing artifacts and noise amplification. This study proposes an optimized reconstruction framework that jointly models coil sensitivity encoding, Wave-CAIPI three-dimensional aliasing dispersion, and sparse regularization to obtain stable solutions under highly underdetermined conditions. Methods Based on the sparsity-regularized Wave-SNMs framework, we propose an optimized reconstruction scheme for highly accelerated brain MRI with multi-coil arrays. By analyzing the statistical properties of k-space data under the complex Gaussian distribution assumption and incorporating the Wave-CAIPI sampling strategy, we investigate the statistical distribution characteristics of highly undersampled MRI data and the sparsity properties of Wave-CAIPI sampling. Accordingly, an L1-regularized Wave-SNMs model is constructed, which uses sparsity constraints to optimize multi-coil image reconstruction. Results Under the 12-fold accelerated MULTIPLEX scanning mode on a 3 T scanner (Siemens Prisma) with a 32-channel head coil using T1-weighted MP-RAGE and multi-echo GRE sequences, the proposed method achieves high consistency with fully sampled reference scans in quantitative assessments of T1, T2*, R2*, and magnetic susceptibility parameters, while effectively suppressing aliasing artifacts and noise amplification. Conclusions Combining statistical distribution modeling with L1 sparsity-regularized optimization effectively improves the accuracy and robustness of high-fold accelerated MRI reconstruction, providing a stable and interpretable reconstruction framework for multi-coil highly undersampled data.