Study On Neural Spike Sorting And Neural Coding

Simultaneous recording of multielectrode array makes it possible to study neural coding or modeling of the nervous system, and it proves to be a useful neurophysiologial measurements. However, it also provides challenging problem of analyzing the recording data. One of factor is that, a micro-electrode usually picks up the activity of multiple neurons in the same region. The other factor is the electrical interference caused by neural interactions, changes in membrane ion channels, and changes in neurotransmitter parameters. The task that accurately assigns each spike to individual neuron, is termed spike sorting, which present important challenges for subsequent analyses, such as neural coding.Generally, heavy noise degrades the performance of many spike sorting methods. In order to receive higher accuracy of spike sorting, we study the effect to performance when applying different mother wavelets to wavelet transform, the performance between wavelet analysis and PCA analysis is compared; A novel spike sorting framework using wavelet features and dynamic mixture of Gaussians clustering method is proposed; On the basis of studying spike sorting, the role of spike trains during neural coding stage is studied. The main contributions of this thesis are summarized as follows:First,we review various techniques of spike sorting and neural coding. Each method is investigated and analyzed in detail. The advantages and disadvantages of these methods are summarized.Then, we study the effect to spike sorting performance when applying different mother wavelets to wavelet transform. Wavelet transform is employed to extract spatial and temporal features which represent spike signal in a more separable way. The experiments on simulated spike signals show that sym5 mother wavelet performs very well even for the heavy noise data, which outperforms other mother wavelet. So we draw the conclusion that, it is more appropriate to apply sym5 mother wavelet to the spike sorting issue.Furthermore, we introduce a novel spike sorting algorithm framework based on wavelet feature and dynamic mixture-of-Gaussians clustering. After spike detection using amplitude threshold method, sym5 wavelet is employed to extract features representing spikes generated by different source neuron from both time and frequency domain. Considering the non-stationary nature of spike trains data, the wavelet feature is divided into short time frames. Then, the dynamic clustering process proceeds in a Bayesian framework, with the source neurons modeled as Gaussian mixtures. The experiments on simulated data demonstrate that our spike sorting method achieves encouraging misclassified rate, better robustness and reliability. Moreover, experiments on real spike data show that the clustering results highly agree with human sorter.Finally, we present an advanced Lempel-Ziv complexity. This new measurement advances the way of symbolizing spike trains. Different partition method of state space is discussed, and then the partition method is extent to the general case, which focus on the fluctuation of the neural dynamic systems. The presented method proves to be a valid feature which solves the problem of coarse-graining. Experiments of visual cortex firing show that, in comparison with standard complexity, our partition complexity representing the response to various stimulus is quite separable. Especially, generating partition complexity performs best on representing the characteristic of spike trains.In all, in this thesis, we not only propose a novel spike sorting method improving the sorting performance, but also study the stimulus-response behavior via partition complexity. This study lays a foundation for further research of neural coding.