Spike Sorting of Microelectrode Single-channel Recordings

Abstract

Until now the experimental research in medical neuroscience has been limited to analyzing summary activity of large neuron populations. However, thanks to recent e orts in using neuroinformatics, arti cial intelligence and machine learning methods in neuroscience as well as medical and technological advances, new opportunities to record activity of individual neurons arises. These opportunities allow us to better understand the neural mechanism of complex behavior, as well as identify parts of the brain responsible for speci c tasks. In this thesis, we were focusing on applying such methods to data recorded from patients with Parkinson's disease that were treated with deep brain stimulation, to improve our understanding of the human brain and the mechanism of the deep brain stimulation in particular. This thesis concentrated mainly on two problems in this eld. First, the evaluation of the state-of-the-art methods used to identify and classify neuronal action potentials (i.e spike sorting methods) in microelectrode recordings, which required devising and implementation of signal generator that produced arti cial signals with similar properties as the signals recorded from basal ganglia. Second, to use these methods to discriminate individual neurons from a microelectrode recording and to use this knowledge to identify neurons with speci c functions in basal ganglia and better understanding of the human brain in general. Spike sorting methods allowed us to nd approx. 20% of basal ganglia neurons with activity related to control of eye movements and 17% of basal ganglia neurons with activity related to processing emotional stimuli or responding to di erent types of emotional stimuli. We were also able to nd several statistically signi cant relations between severity of Parkinson's disease symptoms (described using Uni ed Parkinson's Disease Rating Scale subscores) and statistical characteristics of both microelectrode records and by individual neuron ring patterns using linear mixed-e ects models.