Meeting Banner
Abstract #2291

Searchlight Goes GPU - Fast Multi-Voxel Pattern Analysis of fMRI Data

Anders Eklund1, Malin Bjrnsdotter2, 3, Johannes Stelzer4, Stephen LaConte1, 5

1Virginia Tech Carilion Research Institute, Roanoke, VA, United States; 2University of Gothenburg, Gteborg, Sweden; 3Nanyang Technological University, Singapore, Singapore; 4Max-Planck-Institute for Human Cognitive and Brain Sciences, Leipzig, Germany; 5School of Biomedical Engineering & Sciences, Virginia Tech-Wake Forest University, Blacksburg, VA, United States

The searchlight algorithm is a popular choice for locally-multivariate decoding of fMRI data. A substantial drawback of searchlight is the increase in computational complexity, compared to the univariate general linear model. This is especially true for large searchlight spheres, non-linear classifiers, cross validation schemes and statistical permutation testing. Here we therefore present a graphics processing unit (GPU) implementation of the searchlight algorithm, to enable fast locally-multivariate fMRI analysis. The GPU implementation is 21 times faster than a multithreaded Matlab implementation. This makes it possible to apply 10 000 permutations with leave-one-out cross-validation in about 19 minutes.

Abstract PDF Poster PDF

Keywords

accelerate according accuracy activity advanced affordable another applications applied applying binomial biomedical block brain calculations card cards challenges choice classification classifier classify cluster computation computationally computer computing control core cores correct corrected correction cross cubic dataset datasets days decoding demanding design discriminate distributed distribution done drawback efficient empirical engineering especially examples exchanged fast faster field functional future goes graphics hours human hypothesis implement implementation implementations institute journal labels lack language layer leave linear locally make mapping marked measure medical model motion motor multiprocessing multivariate neighborhood neural nonparametric null overly packages parallel parallelized parametric pattern performance periods permissive permutation permutations popular possibilities power primer prior problems processing programming programs random reconstruction reconstructions reliable repetition requires rest saved scanner schemes school sciences search searchlight simple software speedup spheres statistical straightforward substantial support table tech technological temporal thread threshold thresholded today trained true turner typically valid validated validation vector volume volumes whitening wide wise