Neural Electrical Signal Acquisition And Clustering Analysis Of Signal Sources

Recent years,analysis of brain structure provides a strong foundation for artificial intelligence research,the study and application in artificial intelligence has long become an important way for all walks of life to move towards more cutting-edge development.However,understand the brain structure more comprehensive,and analyze the intricate process of human brain from perception to cognition to judgment until to feedback,is the precondition of making artificial intelligence become the real "intelligent".The development of BCI is an important part to achieve this goal.Among them,how to efficiently and truly obtain the neural signal data of the brain and how to analyze the data to get useful information are the necessary research contents to explore brain science and brain-like intelligence,have great significance for understanding the firing structure of neurons and studying the construction of brain neural networks.In this paper,a set of high-speed acquisition and offline processing of neural electrical signals system is designed,to complete the acquisition of neural electrical signal and the research of neural electrical signal spike clustering algorithm.The part of neural electrical signal acquisition,achieved a set of neural electrical signal collecting system,STM32F407ZGT6 is used as the main control chip and RHD2132 invasive neuroelectrical signal acquisition chip is used as the front end of simulation,LAN8720A as the external Ethernet PHY chip,continuously collect Local neural electrical signals in mice(LFP)for a period of time and save them offline.According to the data characteristics of large amount,high frequency of sampling and large data stream bandwidth,use DMA double buffer mode to achieve data upload while reading at high speed,send signal to the host computer and save it in Excel format,complete the collection of neural electrical signals.The part of neural electrical signal spike clustering algorithm research:Because a piece of brain tissue is collected when the acquisition chip was implanted into the brain of a mouse,in which different nerve cells fire together,in this case the signal sequence collected is the superposition of electrical signals of multiple cells.So that,Spike clustering is used to classify the spikes from the same signal source(neuron)into one catagory,to obtain the number of firing neurons and the firing waveform.In this paper,three sections of work will be complete based on spike clustering technology.1)Select the amplitude threshold method to calculate the threshold for the spikes extraction in the acquisition channel.2)Use the PCA feature analysis method to extract the first three principal components of each spike,realize dimensionality reduction of data.3)Perform cluster analysis on the each spike,use separately four clustering algorithm include K-maens、FCM、KFCM and RKFCM,among them,RKFCM(Robust Gaussian Kernel Fuzzy C-Means Clustering Algorithm)is an improvement of the spike clustering algorithm in this article.108 sets of experimental data were constructed for algorithm verification,to compare the accuracy and reliability of different clustering algorithms in peak clustering.Calculate the average value of partition coefficient(VPC)and partition entropy(VPE)of 108 sets of data,it was found that the improved RKFCM clustering algorithm has an obviously promote in the VPC performance,It proves that the RKFCM algorithm has higher reliability in distinguishing different spikes.Finally,a spike clustering process is showed,use the RKFCM algorithm to perform spike clust analysis on the data of one of the channels collected by the neuroelectric signal acquisition system,the waveform characteristics of the obtained spikes are obvious and the shape is clear.