| Algorithms are presented for rapid, fully automatic, two-dimensional (2-D) and three-dimensional (3-D) tracing of neurons that are imaged by fluorescence confocal microscopy. The speed, robustness and accuracy of these algorithms make them suitable for compelling large-scale applications such as the Human Brain Project, high-throughput neuro-toxicology assays, angiogenesis studies for cancer research, and quantitation of the growth patterns of neurons on nano-fabricated surfaces.; The algorithms work by recursively following the neuronal topology, guided by the correlation response of a set of 4 × N 2 directional kernels in the 3-D case, and a set of 2 × N directional kernels in the 2-D case (e.g. N = 32). These kernels are derived based on a generalized 3-D/2-D cylinder model of the neuronal structures.; This thesis significantly extends prior work on recursive vectorization in several ways. First, the prior work has been extended to trace 3-D structures from fully three-dimensional (volumetric) images captured by 3-D microscopy. Furthermore, the algorithm operates directly on raw confocal image stacks without the need for expensive pre-processing steps such as deconvolution. Second, the prior work has been extended to account for the noise and artifacts that are specific to confacal microscopy of dye-injected neurons. These images are characterized by photon noise, varying contrast, and apparent discontinuity and/or localized hollowness of the structures of interest. It is shown that these barriers can be overcome with a combination of novel algorithmic innovations.; The end product is a labeling of all somas present, graph-theoretic representations of all dendritic/axonal structures, and important statistics such as soma area/volumes, soma centroids, soma interconnectivity, lengths and geometric descriptions of the dendritic trees originating from a soma. The output is generated in a publicly adopted format for ready dissemination within the neurobiology community.; Experimental results are presented to demonstrate the effectiveness, accuracy, consistency, and validity of the proposed algorithms. Studies were also performed to demonstrate the consistency of the tracing algorithms to linear and non-linear image transformations. |
