Accelerating Fibre Orientation Estimation from Diffusion Weighted Magnetic Resonance Imaging Using GPUs

• DOI: 10.1371/journal.pone.0061892
• Publication Date: 2013-04-29T21:09:29Z
• Authors: Moisés Hernández; Ginés D. Guerrero; José M. Cecilia; José M. García; Alberto Inuggi; Saad Jbabdi; Timothy E. J. Beh...; Stamatios N. Soti...
• Keywords: Graphics processing unit

With the performance of central processing units (CPUs) having effectively reached a limit, parallel offers an alternative for applications with high computational demands. Modern graphics (GPUs) are massively processors that can execute simultaneously thousands light-weight processes. In this study, we propose and implement GPU-based design popular method is used analysis brain magnetic resonance imaging (MRI). More specifically, concerned model-based approach extracting tissue structural information from diffusion-weighted (DW) MRI data. DW-MRI offers, through tractography approaches, only way to study connectivity, non-invasively in-vivo. We parallelise Bayesian inference framework ball & stick model, as it implemented in toolbox FSL software package (University Oxford). For our implementation, utilise Compute Unified Device Architecture (CUDA) programming model. show parameter estimation, performed Markov Chain Monte Carlo (MCMC), accelerated by at least two orders magnitude, when comparing single GPU respective sequential single-core CPU version. also illustrate similar speed-up factors (up 120x) multi-GPU multi-CPU implementation.