BROCCOLI: Software for fast fMRI analysis on many-core CPUs and GPUs

Abstract

Analysis of functional magnetic resonance imaging (fMRI) data is becoming ever more computationally demanding as temporal and spatial resolutions improve, and large, publicly available data sets proliferate. Moreover, methodological improvements in the neuroimaging pipeline, such as non-linear spatial normalization, non-parametric permutation tests and Bayesian Markov Chain Monte Carlo approaches, can dramatically increase the computational burden. Despite these challenges, there do not yet exist any fMRI software packages which leverage inexpensive and powerful graphics processing units (GPUs) to perform these analyses. Here, we therefore present BROCCOLI, a free software package written in OpenCL (Open Computing Language) that can be used for parallel analysis of fMRI data on a large variety of hardware configurations. BROCCOLI has, for example, been tested with an Intel CPU, an Nvidia GPU, and an AMD GPU. These tests show that parallel processing of fMRI data can lead to significantly faster analysis pipelines. This speedup can be achieved on relatively standard hardware, but further, dramatic speed improvements require only a modest investment in GPU hardware. BROCCOLI (running on a GPU) can perform non-linear spatial normalization to a 1 mm3 brain template in 4–6 s, and run a second level permutation test with 10,000 permutations in about a minute. These non-parametric tests are generally more robust than their parametric counterparts, and can also enable more sophisticated analyses by estimating complicated null distributions. Additionally, BROCCOLI includes support for Bayesian first-level fMRI analysis using a Gibbs sampler. The new software is freely available under GNU GPL3 and can be downloaded from github (https://github.com/wanderine/BROCCOLI/).

Cluster failure: Why fMRI inferences for spatial extent have inflated false-positive rates

“…The main drawback of a permutation test is the increase in computational complexity, as the group analysis needs to be repeated 1,000-10,000 times. However, this increased processing time is not a problem in practice, as for typical sample sizes a desktop computer can run a permutation test for neuroimaging data in less than a minute (27,43). Although we note that metaanalysis can play an important role in teasing apart false-positive findings from consistent results, that does not mitigate the need for accurate inferential tools that give valid results for each and every study.…”

“…The ordinary least-squares (OLS) functions only use the parameter estimates of BOLD response magnitude from each subject in the group analysis, whereas FLAME1 in FSL and 3dMEMA in AFNI also consider the variance of the subjectspecific parameter estimates. To compare the parametric statistical methods used by SPM, FSL, and AFNI to a nonparametric method, all analyses were also performed using a permutation test (22,23,27). All tools were used to generate inferences corrected for the FWE rate over the whole brain.…”

Empirically investigating the statistical validity of SPM, FSL and AFNI for single subject fMRI analysis

“…methods have become more flexible [15], that multivariate statistical tests have more complicated null distributions than univariate tests [16,17,18] and that more computing power is available [19,20], it may be time for the fMRI community to consider non-parametric statistical methods. A permutation test, for example, does not assume Gaussian data, a constant noise variance or a constant smoothness (and the smoothness does not need to be estimated from the data).…”

Evaluating methods of correcting for multiple comparisons implemented in SPM12 in social neuroscience fMRI studies: an example from moral psychology

“…The application of SnPM is considered less stringent than the Bonferroni and RFT-FWE correction methods applied by software supporting parametric analysis when it is applied using voxel-wise inference (Eklund et al, 2016;Nichols & Hayasaka, 2003); however, clusterwise inference with SnPM is considered to be more stringent (Eklund et al, 2016). Furthermore, the randomise function in FSL and the corresponding function in the BROCCOLI software (Eklund, Dufort, Villani, & LaConte, 2014), which are based on the same statistical principles used for SnPM, have been utilized to evaluate falsepositive rates and sensitivity of traditional correction methods (Eklund et al, 2016).Although clusterwise inference has become a popular method for multiple comparison correction, a recent study evaluating different correction methods has raised concerns that the RFT-applied FWE correction method for clusterwise inference implemented in widely-used fMRI analysis software, such as SPM and FSL, inflates false-positive rates and produces erroneous outcomes (Eklund et al, 2016). RFT clusterwise inference relies on two strong assumptions that might cause such erroneous outcomes.…”

“…The application of SnPM is considered less stringent than the Bonferroni and RFT-FWE correction methods applied by software supporting parametric analysis when it is applied using voxel-wise inference (Eklund et al, 2016;Nichols & Hayasaka, 2003); however, clusterwise inference with SnPM is considered to be more stringent (Eklund et al, 2016). Furthermore, the randomise function in FSL and the corresponding function in the BROCCOLI software (Eklund, Dufort, Villani, & LaConte, 2014), which are based on the same statistical principles used for SnPM, have been utilized to evaluate falsepositive rates and sensitivity of traditional correction methods (Eklund et al, 2016).…”