INRIAlign toolbox for fMRI realignment in SPM99

The INRIAlign toolbox enhances the standard SPM realignment routine (see topic: spm_realign_ui in SPM99 documentation). In the latter, rigid registration is achieved by minimization of the sum of squared intensity differences (SSD) between two images. As noted by several SPM users, SSD based registration may be biased by a variety of image artifacts and also by activated areas. To get around this problem, INRIAlign reduces the influence of large intensity differences by weighting errors using a non-quadratic, slowly-increasing function (rho function). This is basically the principle of an M-estimator.

When launching INRIAlign, the user may select a specific rho function as well as an associated relative cut-off distance (which is needed by most of the rho functions). By default, the rho function is that of Geman-McClure while the relative cut-off distance is set to 2.5.

Apart from this distinction, the method is very similar to spm_realign and uses the same editable default parameters. Most of the implementation has been directly adapted from the code written by J. Ashburner.

Author: Alexis Roche, INRIA Sophia Antipolis, EPIDAURE Group, Now working at the CEA anatomo-fonctional neuro-imaging unit, Frederic Joliot Hospital, Orsay, France

References

L. Freire, A. Roche and J.-Fr. Mangin. What is the best similarity measure for motion correction in fMRI? IEEE Transactions in Medical Imaging 21, p. 470-484, 2002.
L. Freire and J.-F. Mangin. Motion correction algorithms may create spurious brain activations in the absence of subject motion. Neuroimage 14(3), p. 709-722, september 2001.
P.J. Rousseeuw and A.M. Leroy. Robust Regression and Outlier Detection. Wiley Series in Probability and Mathematical Statistics. 1987.

Matlab routines

Xavier Pennec

Last modified: Thu Oct 14 18:47:06 MEST 2004