M.S. (Master of Science)
Department of Computer Science
Imaging systems--Image quality; Three-dimensional imaging; Automatic machinery--Computer programs
This thesis explores computational methods for automatically detecting and quantifying synapses in complex 3D neuronal images. The resulting approach is a novel combination of traditional image processing, machine learning algorithms, and multi-channel comparison methods designed to overcome the unique challenges posed by these images. The methods investigated combine the strengths of each of these components in order to produce an overall method that is capable of fully detecting the synapses in large 3D confocal neuron images with minimal interaction. Human annotation of 3D neuron images remains prohibitively difficult and subjective, and computational analysis tools are highly desirable in the expanding field of 3D neuronal imaging. Validation techniques were also designed and implemented in order to test these methods for this thesis, including construction of a gold standard set of manually annotated synapse images. These are unique in their own right as there are currently no other data sets available for comparison. These methods were tested on multiple partial dendritic tree 3D images and a complete 3D dendrite with good outcomes. Quantitative validation was performed using the gold standard set to check the accuracy of synapse quantification, also with favorable results.
Sanders, Jonathan William, "Automatic machine learning-guided methods for 3D synapse quantification in confocal neuron images" (2016). Graduate Research Theses & Dissertations. 1596.
vi, 65 pages
Northern Illinois University
Rights Statement 2
NIU theses are protected by copyright. They may be viewed from Huskie Commons for any purpose, but reproduction or distribution in any format is prohibited without the written permission of the authors.