Font Size: a A A

Spike Based Brain-machine Interface Algorithm Research

Posted on:2012-04-19Degree:MasterType:Thesis
Country:ChinaCandidate:X C LiuFull Text:PDF
GTID:2178330332478537Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
Brain-machine interface (BMI) provides an alternate communication channel linking directly the nervous system with man-made devices. Through anlyzing and modeling of the neural signals, BMI enables direct interaction and control between the brain and external devices. Because spike signals contain large amount of information and have relatively high temporal and spatial resolution, the spike based BMI is a promising technique in building real-time and accurate prosthetic devices. Therefore, this paper mainly focuses on spike based BMI.This paper studies four critical problems in spike based BMI:spike classification, neural decoding, on-line neural decoding and controlling system and neural information reduction.As a vital preprocessing procedure, first, this paper builds a spike classifier in the form of a rule-set. This classifier gets relatively high accuracy and provides intelligible knowledge about sorting of neural spikes, while in the artificial neural network methods the sorting spike sorting ability is hidden in the weights and structures of networks. Neurophysiology researchers can supply physiological interpretation of the sorting knowledge.Second, this paper introduces a novel continuous neural decoding method based on general regression neural network (GRNN). We compared the performance of the GRNN model with several commonly used methods including the Wiener filter model, the Kalman filter model and the feed forward back propagation neural network (FFBP) model in decoding rats'neural activity of Motor Cortex (M1) during rats'lever pressing task. Experiments show that GRNN has superior capacity for neural decoding.After that, we build an on-line decoding and controling system. The system intergrates several procuderes including signal recording, preprocessing, neural decoding and controling. Using this system, the rat can directly control a mechancal device with its brain signals. At last, we try to provide a solution for redundancy in neural information which results in lower generalization and higher computational consumption. We propose a rough set based method to evaluate the importance of the neurons in neural decoding. Using this evaluation, we build a subset of the neural information which is mathematically sufficient and necessary for decoding the target. Using this subset, we get similar decoding accuracy with much less computational consumption.
Keywords/Search Tags:brain-machine interface, neural decoding, spike sorting, neural information reduction, rough set, general regression neural network
PDF Full Text Request
Related items