Multiscale Cortical Parcellation Based On Geodesic Distance And Hierarchical Clustering

Brain neuronal networks of structural and func-tional connections have a hierarchical organization and a complex relationship between them. To study brain dynamics, it is important to identify the cortical level of parcellation of greater metastability. This paper presents a new multiscale cortical parcellation method based on the geodesic distance between vertices of the cortical surface and agglomerative hierarchical clustering, starting from an anatomical parcellation. First, the centroids of each region are efficiently calculated using the geodesic distance between the region's vertices. Then, an affinity graph is constructed between the region centroids, based on the geodesic distance, from which a dendrogram is constructed using hierarchical clustering. Finally, an adaptive tree partitioning method is employed to obtain parcellations at various granularity levels, producing a multiscale parcellation. Furthermore, we propose an optimized method for the calculation of structural connectomes for each parcellation level. This framework will be made available and can be applied to different fine-grained parcellations. Additional information, such as structural connectivity information can be easily added to the framework. In future work this multiscale cortical parcellation will allow for simulations of cerebral dynamics at different levels.