Hu Cheng1, Andrea Koenigsberger1, Sharlene Newman1, and Olaf Sporns1
Random parcellations have some advantages over
template-based parcellations in network analysis of the brain. An important criterion for assessing the
“goodness” of a random parcellation is the parcel size variability. A new algorithm is proposed to create more homogeneous random parcellations than previously reported. The new algorithm takes the actual distance between voxels and local voxel density into account in placing the random seeds. With many random parcellations using our approach, global network properties exhibit normal distribution and the variability across
different repetitions of the random parcellation
is comparable with inter-subject variability.