Research On Spike Sorting And Frequency Characteristics Of Neuron Network

A large number of neuronal spike discharge sequence is available because of the multi-electrodearrays (MEAs) technology. However, the recording obtained by MEAs is commonly thesuperposition of the spike trains produced by a number of neurons in the vicinity of the electrode.In order to obtain neural signals decoding of spike trains, the neural signals collected by the MEAswas required. Spike-sorting which is important premise of analyses neural coding.Firstly, the paper based on the spike in the time-frequency domain characteristics, analysis theinfluence of wavelet base function to spike-sorting. Proposed a multi-wavelet fusion featurealgorithm for spike extract features, multi-wavelet feature and KS test to achieve featuredimension reduction. Then, based on the characteristics of the spike waveform are non-stationaryand non-linear, the waveform amplitude features are extracted by using the signal gradient ofpermutation entropy method and valley-seeking clustering algorithm, analysis the information andcomplexity of non-homologous spike waveform characteristics，complete realize the effectivespike-sorting.The experiment results showed that the two methods complete spike-sorting,according to spike characteristics departure have some validity. Based on the pattern classificationof neurons, according to the Hodgkin-Huxley (H-H) model builds the feedback delay mechanism.First analysis the processing mechanism based on single H-H model, and then, using double-layerH-H neural network model for further study.The main contributions of this thesis are summarized as follows:(1) Briefly state the neurons spike patterns classification and the nerve information encode ofsignificance, analysis the existing spike-sorting method and information encoding mechanism,finally summarize the validity and limitations of them.(2) Considering the drawbacks of the traditional wavelet transform, such as the subjectiveselection for basic functions, we proposed a novel multi-wavelet fusion feature algorithm for spikesorting. According to the complementary differences of various wavelet basic functions, thedescription of spikes characteristics in the time-frequency domain was completed. The experimentresults showed that multi-wavelet fusion method has a better performance compared with thetraditional wavelet transform，When using sym5 and bior2.4 as the wavelet fusion functions, themisclassification rate of spikes was about 0.65% ~1.17%. The multi-wavelet fusion feature canimprove the objectivity, and has a better feasibility in spike sorting.(3) According to nonlinearity feature of spike waveform amplitude, in order to characterize theinformation and complexity difference among different types of spike waveform, we propose a novel Signal Gradient of Permutation Entropy feature algorithm for spike sorting. Furthermore, weachieve the clustering of three-dimensional feature of the non-homologous spike by usingvalley-seeking clustering. The experiment results showed that the spike sorting method that usingSignal Gradient Permutation Entropy features with valley-seeking clustering，the misclassificationrate of spikes was about 0.78% ~1.34%，Compared with the permutation entropy，the feature ofWithin-class variance is Smaller、between-class variance is large and stable，and can perfectlydistinguish the interference of non- homologous spikes.(4) According to the original Hodgkin-Huxley (H-H) neuron model. Promote the existencefeedback strength mechanism of the H-H model，analysis of different feedback parameters affectthe frequency stability of the spike.Then double-layer H-H neural network model is built,analysisdifferent input stimulation frequency how affect spike output frequency. On the basis of a singleinput, analysis a specific input stimulus effects of spike output; Foe example of two-input caseshows that impact of the spike output frequency. The experimental analysis shows that: to spikeoutput frequency, the high-frequency input stimulation plays a leading role; at the same time, in acertain frequency range of stimulus,to spike output frequency, the high-frequency input stimulationwith low-frequency input stimulation plays a promote role. The research shows that frequencycoding may be a viable mode of neural coding.