The Overlapping Spike Sorting Based On Sparse Coding And Compressive Sensing

Spike sorting is of great significance for understanding brain activity.For example,hundreds of microelectrode arrays that can be implanted in the body are expected to provide promise for fundamental neuroscience research related to neurasthenia.In general,spike sorting is a process of obtaining useful information from an extracellular recording data.With the development of neuronal spike acquisition technology,thousands of neuronal spike signals can be collected at the same time.In this case,each neuron may produce a spike signal by firing at the same time,and the recording data of the electrode will include an overlapping spike signal.How to solve the overlapping spike sorting is a valuable research issue.The existing clustering method has the advantage of higher accuracy and lower computational complexity in spike sorting,but when there is overlapping spike phenomenon,the clustering method is prone to produce higher error rate.Continuous Basis Pursuit(CBP)algorithm can solve overlapping spike sorting problem better,but when the spike waveforms generated by different neurons are more similar,it is difficult to perform spike sorting correctly.In view of this,this paper adopts sparse coding and compressed sensing algorithms to solve the overlapping spike sorting problem,respectively.First,the overlapping spike production process is analyzed by the convolution model.Secondly,the spike template of each neuron can be obtained through the centers of the cluster.The Toeplitz matrix in the convolution model can be obtained by different time-shifted spike template.Then sparse coding and compressed sensing algorithm are used to solve the sparse signal in the convolution model,and the sparse signal is searched by the maximum a posteriori estimation method,so as to achieve spike sorting.In order to verify the effectiveness and feasibility of the sparse coding and compressive sensing algorithm,we selected three sets of simulation data and two sets of measured data in the experiment to verify experimental results.The experimental results show that compared with k-means clustering and Superparamagnetic Clustering(SPC),the proposed algorithm has a lower error number.In addition,when the spike waveforms in the experimental data are similar,the sparse coding has 96.3% and the compression sensing has 97.3% average sorting detection.Compared with the average sorting detection of 93.3% of CBP,the proposed algorithm has a certain performance improvement.