Statistics

In FSL, the primary tool for statistical analysis of task-based fMRI data is FEAT, which provides an interface for an entire analysis, from pre-processing (brain extraction, motion and slice-timing correction, temporal and spatial filtering, etc), through statistics using the general linear model (GLM), and inference with correction for multiple testing. FEAT can be used for single-session, multi-session, and multi-subject analyses.

While FEAT is a tool that enables complete analyses of task-based fMRI, FSL also provides various other tools that can be used for statistics of other imaging modalities, including resting-state fMRI, diffusion MRI, structural MRI (such as VBM), among others. It also offers tools for statistics that can be integrated into customized analyses. These include:

Tool	Description
Glm (Glm_gui on Mac)	A graphical tool to help create design matrix and contrast files (t and F) which can then be used as input to other tools such as randomise, PALM, and MELODIC, and fsl_regfilt. For an overview of various common study designs, see here.
randomise	A command-line tool that allows inference using permutation testing with imaging data, including correction for multiple testing.
palm	A command-line tool that extends the functionality of randomise for complex designs, including multivariate, repeated measures/longitudinal, as well as sib-pair designs. PALM also has methods for faster inference and additional methods for multiple testing correction.
cluster	A command-line tool that thresholds a statistical map and produces a table indicating the clusters of contiguous voxels that survive that threshold. If information is available on the smoothness of the map, cluster can also provide p-values for each cluster calculated using the random field theory.
fdr	A command-line tool that adjusts a map of uncorrected p-values using the False Discovery Rate.
mm	A command-line tool that models a statistical map as a mixture of Gaussians, thus providing an alternative method for thresholding and finding significant voxels.
swe	A command-line tool that uses the ''Sandwich'' Estimator for modelling and inference with longitudinal/repeated measures data.