Dear all,
I am doing a voxel-wise analysis within the thalamus and ran into an issue that I hope someone can kindly offer some insights into. Briefly, values within each thalamus voxel are structural connectivity values expressed as percentages (ranging between 0 and 1). Because we are focusing on thalamus only, all voxels outside of thalamus are 0. As a necessary step before between-group comparisons, we normalized and smoothed individual thalamus images. However, relative to voxels closer to the center of thalamus, voxels along the border of thalamus are disproportionately reduced in value due to all the 0s in surrounding voxels outside of thalamus (which we believe is a known "edge effect" inherent to smoothing). As the cluster of group difference could be small and close to the border of thalamus, this could create a potential problem for our analyses.
Therefore, we have two questions:
1. Is there a way to deal with this during the process of smoothing?
2. How much will this bias the group-level comparison results if we just follow the default smoothing settings? Would a smaller kernel size or even not smoothing at all be a better choice?
Thank you very much,
Beier Yao
Beier Yao, M.A.
Doctoral Student in Clinical Psychology
Michigan State University
[log in to unmask]
|