Detection of bursts in neuronal spike trains using hidden Semi-Markov point process models

Neurons in vitro and in vivo have epochs of bursting activity in which firing rates are dramatically elevated. Studies show that bursts are reliable information unit and encode crucial features of neurons and neural systems (Crick 1984, Gabbiani et al. 1996, Lisman 1997, Izhikevich et al. 2003), thus are of great study interest. Various methods for detecting bursts in extracellular spike trains have appeared in previous literature. A natural approach is to consider a state-space model, in which there are hidden up (bursting) and down (non-bursting) states, the simplest of which is a hidden Markov model. Bursts, however, tend to follow a more regular pattern than that assumed in a hidden Markov model. In this thesis, we present a new method using hidden semi-Markov models for extracellular burst detection. We demonstrate the improved performance of our method over other existing methods using both simulation data and actual spike train data obtained from a goldfish retinal ganglion cell.;We further apply our method to spike trains which exhibit bursting adaptation, a special bursting pattern. We fit various distributions to the firing processes within bursts to reflect this pattern in our hidden semi-Markov models for better burst detection performance. We then test our method on simulation data and apply it to real spike trains recorded from the internal globes pallidus of patients with Parkinsons disease.