BRAVO: Bootstrap Regression Analysis of Voxelwise Observations (version 2.0)


BRAVO is now in 2.0 mode. We have expanded the functionality to include moderated mediation (on the A pathways) and two step serial mediation.  We've also fixed a few bugs along the way that made the statistical tests more conservative.

BRAVO is a Matlab toolbox for performing simple and nested regression analysis on voxelwise observations in MRI data.  The key feature of BRAVO is the use of "bootstrap" or permutation statistics to estimate statistical significance.  While this toolbox designed with structural MRI data in mind (e.g., fractional anisotropy, voxel based morphometry), these routines can be applied to any voxelwise measure.  BRAVO is constructed to be a stand-alone program with minimal dependencies except for the Matlab statistics toolbox and NifTI data format.

BRAVO can perform multiple types of analyses:
    • Correlation: After regressing out covariate factors, BRAVO will perform either a parametric (Pearson's) or non-parametric (Spearman's) correlation between an external variable and the voxelwise measure.
    • Regression: An ordinary least squares regression on any set of dependent variables and the voxelwise measure.  Allows for including covariate factors to control for in the regression.
    • Mediation: Uses nested regression models to estimate causal mediating pathways between a dependent variable, one or multiple potential mediators, and the voxelwise measure, while controlling for nuisance variables.


BRAVO 2.0 was benchmarked and developed in Matlab R2013a on an Ubuntu 14.04 system with an Intel(R) Xeon(R) processor (4 cores, 3.10GHz) system and 8Gb of RAM.  Minimum recommended system specifications are dual core processor (at least 2Ghz) and 4Gb RAM (depending on dataset size).  All code was tested using DTI and VBM data with a 91x109x91 matrix size, 2mm voxels and up to 141 participants with analysis times ranging from 2-4hrs.  Larger datasets may significantly lengthen processing time and system memory requirements.


Nearly all BRAVO source code was written by Timothy Verstynen, in collaboration with Kirk Erickson, Peter Gianaros, and Andrea Weinstein at the University of Pittsburgh. Jason Steffener at the University of Concordia helped considerably with the development of BRAVO 2.0.  BRAVO also uses some open source software developed by Alexandros Leontitsis (Spearmans correlation code) and Jimmy Shen (NIFTI loader functions).


Beginning with release 2.0, all code written here is distributed under the BSD 3-clause open source license.   Any derivative work must thus fall under this license. 


Email all inquiries and questions to timothyv [at]