Dear John and Thomas, Thank you for your useful suggestion. I normalized my dataset again after realignment with reslicing and the odd signal jumps/drops have now disappeared. This website also helped me understand what was going on. http://ceo1409.ceo.sai.jrc.it:8080/aladine/v1.2/tutorials/geomcorr/nearest/nn.html With best wishes, Kota KATANODA John Ashburner wrote: > I would guess that you have done image realignment without reslicing, and the > spatial normalisation is incorporating the movement parameters. This would > make what is happening analagous to resampling the realigned images using > nearest neighbour interpolation. > > If you translate an image by anything less than 1/2 voxel, then the resampled > images would remain the same if nearest neighbour interpolation is used. At > a 1/2 voxel translation, the resampled image would suddenly be like a 1 voxel > translation. I think that what you are seeing are these sudden jumps. > > Best regards, > -John > > > >> This is a question about nearest neighbour interpolation >> used in the process of normalization. >> >> I normalized a functional dataset with this interpolation >> method. The normalized images look fine, but in several >> voxels, there are strange signal behaviors. >> >> The attached jpeg image shows an example. That is a plot of >> a timecourse of one voxel in the grey matter. The signal >> proceeds with small ups and downs for several time points, >> but around the first one-third of the timecourse, it >> suddenly jumps up, and after that it proceeds normally >> again. >> >> Such an odd signal behavior is seen in many other voxels. >> Some show jump-ups, some show drop-downs, and others show >> both. Those voxels seem to be distributed around the whole >> brain randomly. >> >> I tried normalizing the same data set with the other two >> interpolation methods, sinc and bilinear. In these two cases >> no strange signal behaviors were observed.