Hello,
We are performing a study with two groups, 107 subjects in total, and just as an exploratory analysis, I wanted to determine if kmeans clustering with k=2 would yield the correct group assignments. To do this I first ran a group-level analysis in FEAT to look at the average activation across all subjects. Then, from that .gfeat directory, I took the filtered_func_data.nii.gz file, reduced it to a v x n (v=voxels in brain mask, n=subjects) matrix, and performed kmeans clustering on this matrix. Then, using the corresponding group assignments, I ran another group-level analysis to compare the two groups. The results Im seeing are one group just having significantly larger brain activation across the entire brain compared to the other, and nothing significant for the other group (higher than the first). Furthermore, some of the regions of peak activation tend to be along the interhemispheric fissure and between the cerebellum and visual cortex, leading me to believe this is in some way related to motion. I've tried quite a few motion-correction strategies (ICA-AROMA, 24-parameter motion model + scrubbing, etc.), but this does not appear to change anything. Does anyone have an idea what might be causing this? Is there actually a problem, or is this to be expected and Im making a problem out of nothing?
Thanks in advance for any assistance any one may provide!
|