Neuronal function is intimately related to neuronal shape. The morphologies of dendrites and axons are key parameters in descriptions of neuronal cell types. Dendritic and axonal structure also play a key role in shaping cellular excitability and synaptic integration. Finally, the overlap of axons and dendrites often predict synaptic connectivity, allowing the estimate of neural circuits and their function from neuronal shape. This principle has been used to produce rough drafts of cortical microcircuits in the cat and rat, as well as olfactory circuits in Drosophila.
To visualize dendritic and axonal arbors neurons are typically labeled with markers (e.g. biocytin or GFP) and imaged using brightfield or fluorescence microscopy. Beginning with the work of Ramón y Cajal, generations of neuroscientists have drawn neuronal arbors on paper. With the advent of personal computers it became possible to trace neurons in 3D from sets of 2D images acquired at different depths (i.e. image stacks). This process transforms the voxel-based content of the raw pictures into a set of interconnected vectors or cylinders representing the spatial coordinates and orientation of each branch. The resulting "digital reconstructions" provide a computationally compact but complete description of dendrites and axons, suitable for quantitative morphometric and stereological analysis, compartmental biophysical modeling, and electrophysiological simulations.
Dendritic and axonal arbors are large and extremely complex. As a result, manual digital reconstructions are labor intensive. Simple dendritic trees within individual tissue sections require hours to days of hard work; reconstructing the full axonal arborization of a single projection cell may require several months of labor. This time-consuming process constitutes a critical bottleneck in comparative neuroanatomy and in the analysis of neural circuitry by light microscopy. Computational scientists often express the opinion that the problem is solvable by automating the reconstruction with appropriate algorithms, at least in part. In fact, both commercial companies and academic labs have produced software programs for this purpose, claiming a successful solution of this issue. Nevertheless, most neuroanatomists claim these solutions lack general applicability, and to this day, continue to reconstruct the vast majority of dendritic and axonal arbors manually.