Simplified Reeb Graph as Effective Shape Descriptor for the Striatum

摘要

In this work, we present a novel image and mesh processing pipeline for the computation of simplified Reeb graphs for closed triangle meshes of the human striatum extracted from 3D-T1 weighted MR images. The method uses active contours for computing the mesh partition and the simplified Reeb graph. Experimental results showed that simplified Reeb graphs, as obtained by our pipeline, provide an intrinsic, effective, and stable descriptor of striatal shapes to be used as an automatic tool for inter-subject mesh registration, mesh decomposition, and striatal shapes comparison. Particularly, the nodes of simplified Reeb graphs proved to be robust landmarks to guide the mesh registration. The quality of the inter-subject mesh registration obtained by the use of simplified Reeb graphs slightly outperformed the one obtained by surface-based registration techniques. In addition, we show the stability of the resulting mesh decomposition, and we propose its use as an automatic alternative to the manual sub-segmentation of the striatum. Finally we show some preliminary results on the inter-group comparisons among neuroleptic-naive schizophrenic patients and matched controls.