| Description |
Neuroscientists are developing new imaging techniques and generating large volumes of data in an effort to understand the complex structure of the nervous system. The complexity and size of this data makes human interpretation a labor intensive task. To aid in the analysis, new segmentation techniques for identifying neurons in these feature rich datasets are required. However, the extremely anisotropic resolution of the data makes segmentation and tracking across slices difficult. Furthermore, the thickness of the slices can make the membranes of the neurons hard to identify. Similarly, structures can change significantly from one section to the next due to slice thickness which makes tracking difficult. This thesis presents a complete method for segmenting many neurons at once in two-dimensional (2D) electron microscopy images and reconstructing and visualizing them in three-dimensions (3D). First, we present an advanced method for identifying neuron membranes in 2D, necessary for whole neuron segmentation, using a machine learning approach. The method described uses a series of artificial neural networks (ANNs) in a framework combined with a feature vector that is composed of image and context; intensities sampled over a stencil neighborhood. Several ANNs are applied in series allowing each ANN to use the classification context; provided by the previous network to improve detection accuracy. To improve the membrane detection, we use information from a nonlinear alignment of sequential learned membrane images in a final ANN that improves membrane detection in each section. The final output, the detected membranes, are used to obtain 2D segmentations of all the neurons in an image. We also present a method that constructs 3D neuron representations by formulating the problem of finding paths through sets of sections as an optimal path computation, which applies a cost function to the identification of a cell from one section to the next and solves this optimization problem using Dijkstras algorithm. This basic formulation accounts for variability or inconsistencies between sections and prioritizes cells based on the evidence of their connectivity. Finally, we present a tool that combines these techniques with a visual user interface that enables users to quickly segment whole neurons in large volumes. |
