I did a search for my issues and couldn't find the answer - I apologize if this has been covered before!
I have a set of hi-resolution (1.5mm isotropic) 4D .nii BOLD images of the medial temporal lobes. I am trying to normalize them using DARTEL for a traditional random effects analysis of my group's functional data.
Two issues are occurring. I am using DARTEL in SPM8. The two issues may or may not be related, so I will list them both here.
***If I do NOT resample the voxels during normalization (leave the voxel sizes at 1.5mm isotropic) I experience none of the following issues. It doesn't matter if I use a smoothing kernel or not, the normalization works fine if I leave the voxels in their original dimensions.
When normalizing the images, resampling from 1.5mm to 1mm isotropic, and I don't smooth the data, the BOLD images get pock-marked with holes. The same DARTEL normalized images, resampled from 1.5mm to 1mm with a 3mm smoothing kernel have no holes. My concern with the smoothing, however, is whether the holes are still occurring, but are just getting smoothed over, or whether something different is going on and the holes truly don't exist in the smoothed version. The voxels in "holes" are not NaNs nor are they all zeros, but drastically reduced signal intensities relative to their adjacent voxels.
So if I resample, and don't smooth, my functionals get pockmarked with holes of very small signal intensity. If, however, I do NOT resample during normalization, the BOLDs look fine, even unsmoothed. It is also worth noting that that these images look perfectly fine resampled and unsmoothed using traditional MNI normalization in SPM.
When normalizing the images, resampling from 1.5mm to 1mm isotropic, the images fail to normalize beyond 152 TRs. For example, if the 4D nii image for run 1 of subject 1 has 300 images in it, only 152 of those images survives normalization. Strangely, the same occurs for every run of every subject - always 152 images in. The "frame" of the images beyond #152 still exist - meaning I can display images 153:300 for each subject. But they are empty frames. Interestingly, running a 1st level analysis fails with these data, with SPM complaining the files are too small. Smoothing does not help this problem.
Again, however, if I normalize these images with DARTEL and do NOT resample the voxels, the are normalized correctly in their entirety. And again, the images are normalized correctly using SPM's traditional MNI normalization.
Any idea what is wrong? My data appear to be perfectly fine based on the native-resolution DARTEL normalization success and the lack of issues using traditional SPM normalization.
Thanks for any help!
Cognitive Neuroimaging Laboratory