BHMSMA {BHMSMAfMRI}R Documentation

Bayesian hierarchical multi-subject multiscale analysis of functional MRI data

Description

Performs BHMSMA (Sanyal & Ferreira, 2012) of fMRI data using wavelet based prior that borrows strength across subjects and returns posterior smoothed versions of the fMRI data

Usage

BHMSMA(nsubject, grid, Data, DesignMatrix, TrueCoeff=NULL, analysis, 
wave.family="DaubLeAsymm", filter.number=6, bc="periodic")

Arguments

nsubject

Number of subjects included in the analysis.

grid

The number of voxels in one row (or, one column) of the brain slice of interest. Must be a power of 2. The total number of voxels is grid^2. The maximum grid value for this package is 512.

Data

The data in form of an array with dimension (nsubject,grid,grid,ntime), where ntime is the size of the time series for each voxel.

DesignMatrix

The design matrix used to generate the data.

TrueCoeff

If available, the true GLM coefficients in form of an array with dimension (nsubject,grid,grid). By default, NULL.

analysis

"MSA" or "SSA", depending on whether performing multi-subject analysis or single subject analysis.

wave.family

The family of wavelets to use - "DaubExPhase" or "DaubLeAsymm". Default is "DaubLeAsymm".

filter.number

The number of vanishing moments of the wavelet. By default 6.

bc

The boundary condition to use - "periodic" or "symmetric". Default is "periodic".

Details

The wavelet computations are performed by using R package 'wavethresh'. For details, check wavethresh package help.

Value

A list containing the following.

GLMCoeffStandardized

An array of dimension (nsubject, grid, grid), containing for each subject the standardized GLM coefficients obtained by fitting GLM to the time-series corresponding to the voxels.

GLMEstimatedSE

An array of dimension (nsubject, grid, grid), containing for each subject the estimated standard errors of the GLM coefficients.

WaveletCoefficientMatrix

A matrix of dimension (nsubject, grid^2-1), containing for each subject the wavelet coefficients of all levels stacked together (by the increasing order of resolution level).

hyperparam

A vector containing the estimates of the six hyperparameters.

hyperparamVar

Estimated covariance matrix of the hyperparameters.

pklj.bar

A matrix of dimension (nsubject, grid^2-1), containing the piklj bar values (see Reference for details).

PostMeanWaveletCoeff

A matrix of size (nsubject, grid^2-1), containing for each subject the posterior mean of the wavelet coefficients of all levels stacked together (by the increasing order of resolution level).

GLMcoeffposterior

An array of dimension (nsubject, grid, grid), containing for each subject the posterior means of the standardized GLM coefficients.

MSE

MSE of the posterior estimates of the GLM coefficients, if the true values of the GLM coefficients are available.

Author(s)

Nilotpal Sanyal <nsanyal@stanford.edu>, Marco Ferreira <marf@vt.edu>

References

Sanyal, Nilotpal, and Ferreira, Marco A.R. (2012). Bayesian hierarchical multi-subject multiscale analysis of functional MRI data. Neuroimage, 63, 3, 1519-1531.

Examples

# Should take less than a minute to run
nsubject <- 3
grid <- 8
ntime <- 4
Data <- array(rnorm(3*8*8*4),dim=c(3,8,8,4))
DesignMatrix <- cbind(c(1,0,1,0), c(1,1,1,1))
analysis <- "multi"
BHMSMA.multi <- BHMSMA(nsubject, grid, Data, DesignMatrix, TrueCoeff=NULL, analysis)

[Package BHMSMAfMRI version 1.3 Index]