It might be caused by the size of the voxel. This is especially true when you have large voxel size and the transition area is quite sensitive. The reason is that the peak location in mm is converted from the peak location of voxel, which is the minimum unit in our calculation.
In SPM, the display is even interpolated into more detail resolution. When you zoom in (click Tool, then click Zoom in), you will know what I mean. Basically, you can have many more detail location within a voxel. Those location may even have different intensity value.
The best way to clarify this issue is to test an image with landmarks. I just did the test, and here's the result:
- Take "avg152T1_LR_nifti.nii" from: http://nifti.nimh.nih.gov/nifti-1/data
- Display it with SPM5
- Focus on the cornor of "R", which is in upper right area of the axial view
- The cornor of letter "R" is located from [50.5 12.5 0] to [51.5 13.5 0]
- View it with pure voxel based program: nii=load_nii('avg152T1_LR_nifti.nii');view_nii(nii);
- You will find that the corner should be either [50 12 0] (inside) or [52 14 0] (outside). This is correct since the voxel size is 2mm in "avg152T1_LR_nifti.nii" image.
That is why I suspect that the voxel_size might be the cause of this issue.
Suggestions:
- Using image with more resolution (smaller voxel_size). However, this might bring "out of memory" problem.
- Using image with more resolution, plus a pre-defined brain region (a binary image, only true in the area that you would like to analyze, any voxel outside of this area will be treated as non-brain area, and will not be considered to analyze). However, the result might be different from the one that you run analysis with whole brain area.
So, please try them based on your computational environment.