Spike Sorting In Implanted Brain-Computer Interface Neurons

The exploration and development of neuroscience has largely depended on acquisition devices and the ability to decode neurons.The main goal of developing neuroscience is to discover how information is expressed and transmitted in neurons.One of the most popular techniques for collecting electrophysiological signals from neurons in implanted BrainComputer Interfaces is the use of extracellular electrode arrays.Each electrode of the electrode device can capture spike signals from neurons nearby.The electrophysiological activities of neurons collected by electrodes and need to be processed,and the whole signal processing process is called spike sorting.The precision of all the steps in the spike sorting process has an important effect on the accuracy of all subsequent analyses.Spike sorting mainly consists of four steps: filtering,detection,feature extraction and clustering.The existing difficulties in the sorting process of neurons mainly include the uncertainty of background noise of data collection,low spike amplitude and signal-to-noise ratio(SNR),as well as the similarity of neuron spike shapes.In conclusion,it is urgent to develop a spike sorting algorithm which can not only adapt to low SNR and low amplitude spike data,but also has high clustering accuracy in similar shape spike signals.Based on the above content,this paper introduces two algorithms for the precision of spike sorting.Specific work contents are as follows:1.In order to solve the problems of low SNR and low amplitude in spike detection,this paper proposes a new algorithm based on heuristic threshold.Firstly,the parameters of the elliptic filter are optimized to reduce the attenuation of useful signals and fully preserve the low amplitude spike.Secondly,heuristic threshold detection formula is used for spike data with low SNR.Setting the parameters and tweaked them constantly.It can effectively reduce the noise interference and improve the detection accuracy of spike algorithm.The proposed algorithm is verified by the extracellular analog recording data provided by the Laboratory of Neural Engineering,University of Leicester,England,and the average detection accuracy can reach65.21% in a variety of SNR data.In addition,the experimental results based on the implantable data collected under the extended grasping motion paradigm of monkeys show that the proposed algorithm can be effectively used for spike detection even in the real environment with uncertain background noise.2.In order to solve the problem of low SNR and similar shape spike clustering,this paper proposes a K-Multi-Means spike clustering algorithm based on improved dynamic time warping.The total bregman divergence,which is not easy to be disturbed by noise,is used to improve the dynamic time warping algorithm to form a new distance measure,which is used in K-Multi-Means clustering algorithm for spike clustering.The experimental results of two groups of neurons with low SNR and similar shape show that the algorithm can reduce the missed and false clusters of the data,and improve the accuracy of clustering.Secondly,the proposed algorithm is applied to human temporal lobe and monkey real datasets for clustering verification,and the spikes in the signals are clustered into different categories,which shows that the proposed clustering algorithm can effectively process the real recorded biological neuron signals and lay the foundation for subsequent decoding work.