Magnetic Resonance Brain Imaging
Modeling and Data Analysis Using R

Appendix C: Data, Software, and Hardware Resorces

We shortly describe how the example data can be obtained and how they should be organized in order to reproduce the examples in this book. The R packages needed can be installed from Neuroconductor and CRAN, respectively. Additionally the The example code has been tested using R version 3.5.2 on macOS High Sierra Version 10.13.6 ( 32GB, Intel Core I7, AMD Radeon R9 M395X ) and Ubuntu 16.04 ( 32GB, Intel Core I7, Intel HD Graphics 620). Parts of the code, especially in Chapters~\ref{dwi} and~\ref{qmri} are computationally demanding in both memory usage, mainly due to the size of data objects involved, and computing time, due to the large number of voxel involved in the analysis. To level the latter, MPI and openMP parallelization are used where appropriate. In case of limited main memory we recommend to restrict evaluations to sub-cubes of the data, using the arguments \code{xind}, \code{yind} and \code{zind} in the functions that read the imaging data, or to use more restrictive brain masks. The calls to \pkg{fslr} functions \code{xfibres} and \code{probtrackx} in Chapter~\ref{dwi} were executed on compute servers equipped with extensive main memory. These calculations, running exclusively in FSL, took up to 230GB of memory and 2 weeks computing time on a single core. The resources needed may be significantly reduced using restrictive seed masks.