Dear Tom,

Within the ROI_MNI_V4 volume anatomical labels are coded by discrete numbers. Smoothing will increase the number of non-zero voxels, thus covering "more of the brain", but you will also lose the discrete numbers / smear the labels, making the ROI_MNI_V4 volume uninterpretable. Instead you would have to create separate images for each of the labels, binarize them so that the label voxels are always coded as 1 and then apply some smoothing. Due to the smoothing voxels could no longer be assigned to a single label, so you would have to create a sort of maximum probability map based on the smoothed labels.

> I also tried the Hammers atlas as you suggested and after corregistration
Maybe you can provide some pictures to get an impression of how many voxels you actually lose when going with AAL or the Hammers atlas? In principle I would suggest to go with the atlases the way they are provided and maybe note that the estimates might not be perfectly accurate. But hopefully it's a minor issue anyway?

> i don't think the registration was very good.
Just in case, what do you mean with registration? Both your mwc1 files and the atlas files should be in MNI space, thus no coregistration in the sense of translating/rotating one of the images onto the other is necessary. You would only go with a Coreg: Reslice step to resample the brain atlas images to the voxel resolution / image size of your mwc1 files in case they have different resolutions. But you would not use Coreg: Estimate or "Estimate & Reslice, as this will change the orientation.

> I also noticed a problem with my GM images: most of them have some parts of the meninges.
> Could you suggest me a better way of preprocessing the GM images than the unified segmentation approach in SPM8 and some other
> alternative for the corregistration?
Segmentation in SPM12 should be superior, as the algorithms have been improved, but (parts of the) meninges still tend to be misclassified in some regions like occipital lobe, around the interhemispheric fissure. You could try a higher MRF Parameter and increase the Clean Up setting.

Best,

Helmut