Extraction of partial skeletons for semi-automatic segmentation of cortical structures

Abstract

Volume-of-interest segmentation is an important part in the functional analysis of multimodal data (MRI, PET). The usual approach of manually segmenting structures in single 2D slices is cumbersome and prone to errors. The authors present an extension of a semiautomatic approach, which allows the user to define a volume of interest on a 3D visualization of the cortical surface. The volume of interest is computed with a volume growing approach, in which a partial sulcal skeleton serves as a boundary. A new 3D algorithm computes a partial skeleton that is more compact and complete compared to 2D algorithms applied in three orthogonal orientations. This method allows us to simplify the subsequent postprocessing steps and thus to compute a more precise segmentation.