Dynamic Functional Connectivity K-mean clustering

Hi everyone.

I am doing a Dynamic Functional Connectivity analysis with some task data (the task hemodynamic response was regressed out in preprocessing). I extracted the bold time series from multipĺe ROIs and computed the dynamic FC matrices with fisher z transformation.

I want to perform a k-means cluster with all the matrices of all subjects to study the dynamic FC states and I am using R’s biganalytics::bigkmeans(), but the clusters are the same within participants. For example, with 21 participants and k=5 (i am trying a range of 2-10), all the matrices of one participant will be part of the same cluster, and the same thing happens for the other participants. There seems to be more variance in those matrices between participants than within them. Am I missing something in the preprocessing/analysis pipeline?