A fast Pytorch based GRAPPA implementation for uniformly undersampled k-space

Motivation: Existing open-source GRAPPA implementations, such as pygrappa, are slow in some cases. Goal(s): To develop a high-performance, open-source GRAPPA algorithm for faster reconstruction of uniformly undersampled data. Approach: We implemented PyTorch-based GRAPPA algorithm that can use either CPU or GPU. Results: Tests with 8- and 32-channel MRI data showed that our implementation delivers similar image quality to pygrappa but with significantly reduced runtime. Impact: This approach enables faster MRI reconstructions, making it suitable for many applications.