GRAPPA reconstruction with spatially varying calibration of self-constraint

Purpose: To develop and evaluate a novel method of generalized auto-calibrating partially parallel acquisition (GRAPPA) with spatially varying calibration of self-constraint for parallel magnetic resonance imaging (MRI) reconstruction.

Theory and methods: The conventional GRAPPA independently estimates each missing sample with adjacent acquired data over multiple coils, thereby ignoring correlations inside missing data. Self-constrained methods can exploit correlations inside missing data by imposing linear dependence within full neighborhood kernels and showing improved reconstruction compared with GRAPPA. However, self-constraint kernels are currently calibrated by using auto-calibration signals. Thus, they may be suboptimal for reconstructing outer k-space because of spatially varying correlations. This study proposes a novel GRAPPA method with separate self-constraints (SSC-GRAPPA). In this method, the spatially varying self-constraint coefficients are adaptively calibrated by separately exploiting correlations inside missing and acquired data in the outer k-space. Both phantom and in vivo imaging experiments were conducted with retrospective undersampling to evaluate the performance of the proposed method.

Results: Compared with GRAPPA and self-constrained GRAPPA, the proposed SSC-GRAPPA generates images with reduced artifacts and noise.

Conclusion: The proposed method provides an effective and efficient approach to improve parallel MRI reconstruction, and has potential to benefit routine clinical practice in the future.