 |
 |
> > Hello there; > > In my experiment , The realignmen parameters for one subject follows the > stimulus application pattern. Any suggestions. Should i drop this data?? > > Regards; > > -s madi > Dear madi, this is not an easy one. Generally speaking, even after having performed a proper realignment there is remaining variance which is strongly correlated to the movement parameters. There are two sources from which this variance may originate, i) Imperfect realignment; due to inadequate interpolation model and/or orientation dependent suceptability artefacts. ii) Spin history effects. The rather carefree approach which has been used in SPM this far has been to regress out any variance that correlates with a function of the movement parameters. This works fine when the movements are orthogonal to the task, but when they follow the stimulus application pattern, as in your case, it will destroy the actual signal. There is presently not any good solution to this problem. We have been trying higher order interpolation models (i.e. k-space realignment) without succeeding to remove the variance. In the next version of SPM (beta due in a couple of months) you will have the option to test k-space realignment yourself. There will also be improvements to the covarying out of the variance, making it a little gentler through the use of a MAP scheme. However, a completely satisfactory solution to the problem of task-correlated movements is still to be found. I should add though that in the case of no actual movements (a rather unlikely case) the realignment algorithm will, in the presence of large activations, erroneously find very tiny ~0.01mm movements that are perfectly correlated to the task. If you believe this to be the case (all movements ~0.01mm) you may try to proceed with the analysis without any relignment. Otherwise, I would advise you to realign your data without "adjusting for sampling errors", and to analyse your data while being very wary of "activations" along any edges in the brain. You may also have a go at them with "adjustment for sampling errors", this time being aware that you may be missing actual activations. Unfourtunately you may well find, once you have interrogated your data, that you end up "dropping" this casee befor you are finished. Good Luck Jesper %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%