BrainVISA toolboxes

fMRI toolbox

Our main goal is to ease neuroimaging studies involving both structural and functional modalities and/or large cohorts, for which automated database management is critical.

The fMRI toolbox incorporates original algorithms developed at Neurospin/LNAO, INRIA Saclay/Parietal and partners to perform both mass univariate analyses a la SPM and less conventional multivariate analyses. Each process comes with an intuitive graphical user interface. Part of the underlying computational engine is independent from BrainVISA, and is integrated to the NiPy1 project, an open-source library for Neuroimaging that mixes Python and C/C++ languages. The following image processing methods are currently featured:

• Pre-processings: simultaneous slice timing and motion correction, coregistration and affine normalization using mutual information-like similarity measures.
• Within-subject, multi-session linear model fitting and testing under AR1 noise model with spatially varying temporal correlation [Roche04].
• One-sample and two-sample group analysis u sing two-level hierarchical models and permutation tests [Roche07].
• Single-subject and multi-subject brain parcellation in homogeneous functional regions using Gau ssian mixture models [Thirion06].
• Joint detection-estimation of brain activity using parcel-based nonparametric hemodynamic response function (HRF) and Markov random field models [Makni08].
• Cortical surface-based functional analysis through fMRI data resampling onto a cortical mesh [Operto08], the mesh itself being obtained from the T 1 MRI toolbox. T his process is borrowed from the separate Cortical Surface toolbox, which collects a number of advanced tools for cortical surface parameterization and analysis.

An example of using the toolbox to perform parcel-based detection-estimation for a single subject is illustrated below. Starting from a gyral parcellation of the cortex [Cachia03b] the method estimates a different HRF in each parcel (shown in blue) based on a nonparametric response model. These estimates are compared with the usual canonical HRF, shown in red.

Another usage example is functional cortical surface-based analysis. After projecting the fMRI time courses onto the cortical surface, one may fit a general linear model in each node of the associated mesh. Activations are then detected via a t-test. To ease the interpretation of the results, sulci extracted using the T1 MRI toolbox are overlaid to the functional map.

The fMRI toolbox for BrainVISA is a comprehensive functional image analysis package, which we plan to further develop in the near future.

DOWNLOAD toolbox for Linux
How to install ?