Hierarchical Clustering of Streamtubes

We apply hierarchical clustering methods on streamtubes for visualization and analysis. Streamtubes are integrated in the major eigenvector field of the DTI data set. In a 256 × 256 × 50 data set, our algorithm can generate tens of thousands of streamtubes. It is hard to find features in a dense set of undistinguished tubes. Thus it is important to impose some structural information on the streamtubes for visualization and interpretation purposes. Hierarchical clustering produces a dendrogram that groups objects into different number of clusters in a continuous way. We apply some clustering methods on a set of streamtubes and found that the streamtubes correlating to major neural structures tend to cluster together because of their shape similarities. Also, different distance criteria produce different types of clusters. The dendrogram produced by the hierarchical clustering methods has the potential to be utilized by visualization applications to interactively display the streamtubes at different level-ofdetails.