The main purpose of this thesis is to study spike sorting methods and to develop a new method using Bayesian Non-parametric Models. Spike sorting is the process of detection, feature extraction, and clustering signals recorded from deep brain embedded electrodes.The output of the process isthe neuronal firing pattern. Spike sorting is the first step in each brain information process.The most important methods for detection, feature extraction and clustering in spike sorting are studied in this thesis.In addition, the two main challenges i.e., the existence of overlapped spikes and variations of spike waveforms over time are investigated. This thesis is focused on non-stationary data clustering and proposes a new method to overcome the second challenge.Model-based clustering and especially Bayesian clustering methods have alreadybeen used for this purpose. In these methods, a mixture distribution is considered as probability density function of the data. It means that each data point is produced by one of the mixture components. In the Bayesian point of view, model parameters are random variables with a prior probability distribution. Bayesian clustering is categorized in two main groups, parametric and non-parametric. In the parametric approaches the number of clusters is assumed to be known and the parameters are unknown. However, the number of clusters is also unknown in the non-parametric Bayesian methods. Dirichlet process mixture (DPM) is one of the methods used for estimating number of clusters. In this thesis a new approach has been developed to track data changes over time based onDPM. Previous frame information is used as the prior for the current frame. Therefore, it is possible to track cluster variations as well as detect changes in the cluster number. Our method was compared with a similar method based on DPM without any prior information. Results showed that the proposed method has a better performance in terms of error rate. Cluster variations were also tracked acceptably. Keywords Spike sorting; Detection, Feature extraction; Clustering; Bayesian inference; Dirichlet proce Non-stationary data
