A new approach is proposed for extracting explicit representations of 3D curvilinear features form stacks of 2D images. The images, which are of brain tissue, were obtained by confocal microscopy and the features represent the dendritic tree structure surrounding a neuron. Voxels with a high probability of being on the center-lines of the dendrites are identified first. Then a combination of a 3D minimum spanning tree and a 3D minimum cost path algorithm is used to automatically extract explicit center-line representations of the curvilinear features. The final objective of the image analysis is to produce, as automatically as possible, generalized cylinder models of the dendritic structures which are then used for studying neuronal morphology and function. In this paper, we concentrate on the algorithms used to extract the center- line representation.
展开▼
