| I have a question: In realignment, what function does the 'adjust sampling
| errors' option have? Does it come into play in the statistics (i.e. removing
| motion correlated activation)?
This is not my favourate piece of work. What it idoes is described in
the help facility for image realignment:
% Adjust the data (fMRI) to remove interpolation errors arising from the
% reslicing of the data. The adjustment for each fMRI session is performed
% independantly of any other session. Bayesian statistics are used to
% attempt to regularize the adjustment in order to prevent an excessive
% amount of signal from being removed. A priori variances for coefficients
% are assumed to be stationary and are estimated by translating the first
% image by a number of different distances using both Fourier and sinc
% interpolation. This gives a ball park figure on how much error is
% likely to arise because of the approximations in the sinc interpolation.
% The certainty of the solution is obtained from the residuals after
% fitting the optimum linear combination of the basis functions through
% the data. Estimates of certainty based on the residuals are
% unfortunately just an approximation.
% We still don't fully understand the nature of the movement artifacts
% that arise using fMRI. The current model is simply attempting to remove
% interpolation errors. There are many other sources of error that the
% model does not attempt to remove.
% It is possible that adjusting the data without taking into account
% the design matrix for the statistics may be problematic when there are
% stimulous correlated movements, since adjusting seperately requires the
% assumption that the movements are independant from the paradigm. It
% MAY BE BE BETTER TO INCLUDE THE ESTIMATED MOTION PARAMETERS AS CONFOUNDS
% WHEN THE STATISTICS ARE RUN. The motion parameters are saved for each
% session, so this should be easily possible.
Best regards,
-John
|