GPU accelerated Cartesian GRAPPA reconstruction using CUDA

•GPU based parallel implementation of GRAPPA for an efficient MR image reconstruction.•The proposed method uses customized CUDA kernels to speed up the CPU-intensive tasks (i.e., weights estimation and interpolations) of the GRAPPA algorithm.•The proposed GPU based implementation of GRAPPA can work for various configuration settings i.e., ACS lines, multiple receiver coils, kernel size and acceleration factors.•The proposed method significantly accelerates the GRAPPA reconstruction time up to a factor of ≈ 47.8×, while maintaining the reconstruction quality.