f.ica.fmri {AnalyzeFMRI}R Documentation

Applies Spatial ICA (Independent Component Analysis) to fMRI datasets


Decomposes an fMRI dataset into a specified number of Spatially Independent Components maps and associated time-courses using the FastICA algorithm


f.ica.fmri(file.name, n.comp, norm.col=TRUE, fun="logcosh", maxit=1000,
alg.type="parallel", alpha=1, tol=1e-04, mask.file.name=NULL, slices=NULL)



path to fMRI dataset (ANALYZE format .img file)


number of components to extract


a logical value indicating whether each voxel time series should be standardised to have zero mean and unit variance before the ICA algorithm is applied (default=TRUE recommended in practice)


the functional form of the G function used in the approximation to negentropy (see details)


maximum number of iterations to perform


if alg.typ=="deflation" the components are extracted one at a time (the default). if alg.typ=="parallel" the components are extracted simultaneously.


constant in range [1,2] used in approximation to negentropy when fun=="logcosh"


a positive scalar giving the tolerance at which the un-mixing matrix is considered to have converged.


Optional path to file containing a 0/1 mask for the dataset


Optional vector of slices to be included


The fMRI dataset is rearranged into a 2-dimensional data matrix X, where the column vectors are voxel time-series. A mask is used to specify which voxels are included. If this is not supplied by the user then a mask is constructed automatically using a 10% intensity threshold.

The data matrix is considered to be a linear combination of non-Gaussian (independent) components i.e. X = AS where rows of S contain the independent components and A is a linear mixing matrix. In short ICA attempts to 'un-mix' the data by estimating an un-mixing matrix U where UX = S.

Under this generative model the measured 'signals' in X will tend to be 'more Gaussian' than the source components (in S) due to the Central Limit Theorem. Thus, in order to extract the independent components/sources we search for an un-mixing matrix U that maximizes the non-gaussianity of the sources.

In FastICA, non-gaussianity is measured using approximations to negentropy (J) which are more robust than kurtosis based measures and fast to compute.

The approximation takes the form

J(y)=[E{G(y)}-E{G(v)}]^2 where v is a N(0,1) r.v

The following choices of G are included as options G(u)=\frac{1}{α} \log \cosh (α u) and G(u)=-\exp(\frac{-u^2}{2})

The FastICA algorithm is used to 'un-mix' the data and recover estimates of the mixing matrix A and the source matrix S. Rows of the source matrix S represent spatially independent components of the dataset (these are arranged spatially in the output). Columns of A contain the associated time-courses of the independent components.

Pre-processing involves removing the mean of each row of the data matrix and (optionally) standardizing the columns of the data matrix to have zero mean and unit variance.

All computations are done using C code. This avoids reading the entire dataset into R and thus saves memory space.


A list containing the following components


estimated mixing matrix


estimated source matrix that has been rearranged spatially i.e. S is a 4-D array and S[,,,i] contains the 3-D map of the ith component


the name of the data file


the name of the mask file


J L Marchini <marchini@stats.ox.ac.uk> and C Heaton <chrisheaton99@yahoo.com>


A. Hyvarinen and E. Oja (2000) Independent Component Analysis: Algorithms and Applications, Neural Networks, 13(4-5):411-430

Beckmann C. (2000) Independent Component Analysis for fMRI. First Year D.Phil Report, Dept. of Engineering Science, University of Oxford.

See Also


[Package AnalyzeFMRI version 1.1-23 Index]