Eelke
Visser1,2, Emil Nijhuis1,3, Marcel P. Zwiers1,2
1Donders Institute for Brain, Cognition
and Behaviour, Radboud University Nijmegen, Nijmegen, Netherlands; 2Department
of Psychiatry, Radboud University Nijmegen Medical Centre, Nijmegen,
Netherlands; 3Department of Technical Medicine, University of
Twente, Enschede, Netherlands
Finding
clusters among the many streamlines produced by tractography algorithms can
improve interpretability and can provide a starting point for further
analysis. A problem with many clustering methods is their handling of large
datasets. We propose to overcome this problem by repeatedly clustering
complementary subselections of streamlines. The execution time of the
algorithm scales linearly with the number of streamlines, while working
memory usage remains constants. The method produces anatomically plausible
and coherent clusters in a single subject. When applied to a large group
dataset, results are similar and consistent across subjects.