Hi Cyril,
> I've an fMRI time series in which there is a missing volume.. (don't
> know why but ..) The idea is to create a new volume - as far as
> I've seen on the list some of us took the two neighbouring volumes
> and computed to mean to get 'intermediate' image. Does anyone knows
> if it's correct or have a better solution?
Instead of inventing/interpolation data, I'd recommend modeling the
full time series with the dud scan, but then nullify its effect with a
dummy regressor. For this approach, you need to specify *some* image
for the missing volume, but it can be anything (as long as it doesn't
mess up the image mask or the global grand mean); if you don't have an
image or are afraid that the bad image will mess up the mask or
global, the replacement image could just be a copy of an adjacent
scan.
To nullify the scan you need to specify extra regressor, all zeros
except for a one at the problem regressor. Below is a function
dummy.m that will produce just such a regressor on the fly.
Hope this helps.
-Tom
-- Thomas Nichols -------------------- Department of Biostatistics
http://www.sph.umich.edu/~nichols University of Michigan
[log in to unmask] 1420 Washington Heights
-------------------------------------- Ann Arbor, MI 48109-2029
function [X] = dummy(n,I)
% Create dummy (indicator) variables
% FORMAT [X] = dummy(n,I)
% n - Number of scans/rows
% I - vector of scans/rows to model
%___________________________________________________________________________
%
% Creates dummy vectors, suitable for entering as additional regressors
% in a SPM analysis.
%
% If I has length p, then X is a n-by-p matrix of all zeros, except
% column k has a one at the I(k)'s row.
%
%___________________________________________________________________________
% $Id: dummy.m,v 1.1 2005/08/18 21:01:11 nichols Exp $
p = length(I);
X = zeros(n,p);
for i = 1:p
X(I(i),i) = 1;
end
return
|