Research On Real-time Spike Sorting Method Based On CUDA

Brain science is a discipline that studies the nature and laws of brain cognition,consciousness,and intelligence.With the emergence of new technologies such as brain imaging,biosensing,human-computer interaction and artificial intelligence,brain science is becoming an important frontier science field with multiple disciplines,and it is also the focus of many countries’ science and technology strategies.An important aspect of brain science research is the study of single neurons and neuron communication problems in the complex construction of neural networks.This research can be used for basic research on the mechanism of action of neurons and early diagnosis and treatment of neurological diseases.On the other hand,it can provide basic theoretical support and model construction schemes for brain research such as neural networks and intelligent robots,and promote its rapid development.Therefore,the study of neurons has become a focus of current brain science research.Spike sorting is one of the effective ways to study neurons.By collecting the neuron spike signal,waveform extraction,feature extraction and clustering,the spike signal of a specific feature can be associated with a specific neuron cell.Together,it is possible to better study the relationship between human behavior,consciousness,and other neuronal cells.With the rapid development of technology,on the one hand,the current high-density electrode probe can simultaneously collect thousands of channels of spike signals.When the collected information is more abundant,the amount of collected data increases exponentially.The traditional spike processing method It takes a lot of processing time and limits the application of real-time scenes.On the other hand,GPU is used for parallel computing from graphics processing,so that large-scale general-purpose computing can be easily parallelized,so that existing algorithms can be parallelized and targeted optimization.Therefore,it is feasible and necessary to design a reliable,real-time,online processing of the spike sorting scheme.The CUDA-based real-time spike sorting proposed in this paper mainly has the following contributions:(1)For real-time high-flux spike data,a real-time spike analysis system framework based on CUDA is proposed,and the system flow is introduced;(2)For the preprocessing of online data,the method of overlapping partitioning is proposed,and the data is divided into overlapping blocks to ensure the waveform integrity at the data block segmentation and improve the detection rate of the spike signal;(3)Analysis of the spike clustering algorithm The parallelism of masked-GMM and parallelization based on CUDA platform;(4)performance analysis of masked-GMM algorithm implemented by parallelization,optimizing the execution efficiency of code to achieve real-time spike sorting;The experiments of collecting and mixing data sets verify the real-time and accuracy of the algorithm,which greatly reduces the limitations of the spike sorting algorithm in real-time application scenarios.