Accelerating Parallel Magnetic Resonance Image Reconstruction on Graphics Processing Units Using CUDA

Magnetic Resonance Imaging (MRI) is a non-invasive and powerful technique for clinical diagnosis and treatment monitoring. However, long data acquisition time in conventional MRI may cause patient discomfort and compliance. Recently, parallel magnetic resonance imaging (pMRI) techniques have been developed to speed-up the MR data acquisition time by collecting a reduced data set (k-space) using multi-channel receiver coils. However, with an increasing number of receiver coils, the handling and processing of a massive MR data limits the performance of pMRI techniques in terms of reconstruction time. Therefore, in real-time clinical settings, high speed systems have become imperative to meet the large data processing requirements of pMRI technique i.e. Generalized Auto-calibrating Partially Parallel Acquisition (GRAPPA). Graphics processing units (GPUs) have recently emerged as a viable solution to adhere the rising demands of fast data processing in pMRI. This work presents the GPU accelerated GRAPPA reconstruction method using optimized CUDA kernels to obtain high-speed reconstructions, where multiple threads simultaneously communicate and cooperate to exploit the fine grained parallelism of GRAPPA reconstruction process. For a fair comparison, the performance of the proposed GPU based GRAPPA reconstruction is evaluated against CPU based GRAPPA. Several experiments against various GRAPPA configuration settings are performed using 8-channel in-vivo 1.5T human head datasets. Experimental results show that the proposed method speeds up the GRAPPA reconstruction time up to 15x without compromising the image quality.