Modelling Neuron Morphology: Automated Reconstruction from Microscopy Images

Understanding how the brain works is, beyond a shadow of doubt, one of the greatest challenges for modern science. Achieving a deep knowledge about the structure, function and development of the nervous system at the molecular, cellular and network levels is crucial in this attempt, as processes at all these scales are intrinsically linked with higher-order cognitive functions. The research in the various areas of neuroscience deals with advanced imaging techniques, collecting an increasing amounts of heterogeneous and complex data at different scales. Then, computational tools and neuroinformatics solutions are required in order to integrate and analyze the massive quantity of acquired information. Within this context, the development of automaticmethods and tools for the study of neuronal anatomy has a central role. The morphological properties of the soma and of the axonal and dendritic arborizations constitute a key discriminant for the neuronal phenotype and play a determinant role in network connectivity. A quantitative analysis allows the study of possible factors influencing neuronal development, the neuropathological abnormalities related to specific syndromes, the relationships between neuronal shape and function, the signal transmission and the network connectivity. Therefore, three-dimensional digital reconstructions of soma, axons and dendrites are indispensable for exploring neural networks. This thesis proposes a novel and completely automatic pipeline for neuron reconstruction with operations ranging from the detection and segmentation of the soma to the dendritic arborization tracing. The pipeline can deal with different datasets and acquisitions both at the network and at the single scale level without any user interventions or manual adjustment. We developed an ad hoc approach for the localization and segmentation of neuron bodies. Then, various methods and research lines have been investigated for the reconstruction of the whole dendritic arborization of each neuron, which is solved both in 2D and in 3D images.