Meeting Banner
Abstract #0688

Dendrogram Processing for Whole-Brain Connectivity-Based Hierarchical Parcellation

David Moreno-Dominguez1, Alfred Anwander1, Thomas R. Knsche1

1Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germany

Hierarchical clustering of probabilistic tractograms encodes the information of the connectivity structure at all granularity levels in a hierarchical tree or dendrogram. It might be the key to whole-brain connectivity based parcellation, where the correct number of clusters is unknown and depends on the desired granularity. The interpretation of the resulting dendrogram is not simple, due to outliers and the size of the dataset encoded, among other reasons. In this study a fast, fully hierarchical bottom-up algorithm is presented, and intelligent processing steps are introduced in order to ease the information extraction process, successfully enabling better performance of tree-partitioning algorithms.

Keywords

account achieved agglomerative among applied arises artifacts atlas available believed best better binary bottom boundaries boundary brain branching centroid challenge cleaned close cluster clustering clusters cognitive cohesion computed connectivity correlated cortex cortical desired developed diffusion distance dominated ease efficiently enable enabled encoded exploration explored extract extraction faced fast find finding flattening form forward foundation frontal full functional furthermore generated granularity great healthy hemisphere hierarchical highest horizontal human in vivo individual inferior inner institute intelligent inter interactively interplay interpretation introduced isolated isolating keeping maximizing might minimizing monotonic must nature neighborhood normalized numerous onto optimal organization outliers outline overcome pair participant particularly partitioning partitions performance poster predefined probabilistic process processing project providing pruning quantify reducing relevant rely researchers resolution restrictions scanner schematic schemes seed selected several sharp similarity since smoothing software solution space step steps strongly structure structures subject subjects suggested suitable surely theory tracking tree trees typically understanding upper usable useful whole wide yielding