Neurons use several methods to integrate incoming information to make a decision about whether to activate an action potential and send information on to other neurons. This process adapts over time and provides the neurons with the ability to learn. One of the forms of learning that is not well understood is branch point plasticity, the ability for different branches in the dendritic arbor to change their ability to conduct information over time. In this project tools are developed to study the structural changes of branch points and indirectly the chemical changes at branch points to better understand the underlying mechanisms behind branch point plasticity. To this end large quantities of neuropil data need to be analyzed. Modern electron microscopy techniques can provide massive quantities of biological image data at extremely high magnification, but the ability to process this data and obtain usable information is a major bottleneck.

Machine learning tools were used to automatically segment cell membranes and mitochondria in large volumes of neuropil electron microscopy data. The new implementation of the algorithms improved their accuracy and efficiency for the given datasets. The algorithms now provide near-human accuracy for organelle and membrane detection at the same speed that data can be acquired from the microscopes using off-the-shelf desktop machines making this tool accessible to labs without specialized computing resources. Specialized programs were then developed to analyze the geometry of the branch points along with the spatial distributions of mitochondria and endoplasmic reticulum relative to the branch points. Using these tools on a sample set of data shows that mitochondria and endoplasmic reticulum volume percentages fluctuate with the distance from the soma. Moving forward, these tools can be used to analyze large datasets to discover the underlying mechanisms of branch point plasticity.