Neuron Morophology Analysis
Implantable neural devices are an emerging technology that has the potential to dramatically change the lives of individuals impaired with ailments such as neural deficits or prosthetic limbs. However, one major barrier to long-term implantation of cortical electrodes is the defense reaction of the human body, which essentially walls-off the electrodes, thus limiting their utility and signal propagation. The design of the eletrodes has a large effect on the reaction of the brain, and this project was aimed at evaluating the response and longetivity of different electrode designs.
This work was part of a large, multi-institution project that included experimental design, animal experimentation, imaging, image processing, and data analysis, where my work on this project included experimental design and data analysis. In terms of the experimental design, I developed a partially-balanced incomplete block (PBIB) design that allowed for evaluating a handful of electrode designs using as few specimens as possible, and the collected data (primarily SNR) could be analyzed with a linear mixed model. Neuron reaction was assessed based on the morphology of the neurons, which were automatically extracted using specially-designs software. These features were used with a variety of clustering methods to categorize the quality of the electrode designs. The most fruitful of these methods proved to be biclustering because of it's intuitive visualization. Additionally, I developed a genetic algorithm for reordering the leaves of a hierarchical clustering tree so that cluster identification becomes visually obvious.