Temporal neurite registration using hierarchical alignments

We study the development of neuronal morphology with respect to temporal correspondences among branching features. We pose the problem as a three-dimensional registration of the observed neuronal morphologies over time, given already calculated neurite traces, where we exploit the inherent tree-like structures. We align branches from pairs of development instances in a hierarchical fashion, from bigger to smaller groups (subtrees), and the alignment costs are accumulated along the defined recursion depths to calculate optimal correspondences. This way, matching is repeated sequentially in different resolutions, first between ‘primary’ branches, and then within their corresponding subtrees. We validate our method using traces of sensory neurons of the larva Drosophila in successive developmental instances (second to third and third to fourth instar).