 |
 |
Thanks a lot for the comments and suggestions, Steve. I based my experimentation only on the fact that the data are smoothed before statistics and thus expected to see some spots. It seems that the asymmetry is probably doing its job. I'll try to go with ROIs. Cheers, Martin On Friday 04 April 2008 09:38:43 Steve Smith wrote: > Hi Martin, > > This probably isn't a great approach as it relies on exact spatial > left-right symmetry in order to be well conditioned. The more normal > approach is to specify L and R ROIs and compare results within them - > this is simpler and doesn't rely on exact spatial symmetry. > > You could for example, generate the ROIs from the standard space > atlases, or maybe from the thresholded activation masks - after L-R > flipping and averaging those masks, etc. etc. > > The measure of activation to compare is something that is not > generally agreed upon - e.g., peak Z, supra-threshold voxel count, > etc. I would recommend reading the Mitsis NeuroImage paper for > investigations of options here. > > Cheers, Steve. > > On 2 Apr 2008, at 20:02, Martin Kavec wrote: > > Hi, > > > > I am trying to test for statistical difference of activation in > > bilateral > > task. I have simple boxcar paradigm, which gives nice activations in > > the both > > L and R hemispheres and I would like to know which one is > > statistically > > higher. I do the following: > > > > 1. flip slice-wise the whole time series so the AP orientation stays > > the same, > > just L->R > > > > 2. merge in time the original and the flipped time series > > > > 3.setup the FEAT with EV1 and EV2, where the EV1 stops after 180s > > (the actual > > length of the the original time series) and the EV2 skips 180s. > > > > 3. set C1 [1, -1] and C2 [-1, 1], though the second one is probably > > not > > necessary > > > > The problem is that I don't get any difference even in the data, > > where the > > analysis of the original time series is evidently asymmetric. I would > > appreciate if anybody could give me a hint what could be wrong with > > this > > approach. > > > > Thanks a lot in advance, > > > > Martin > > --