Research On Brain-computer Interface Based On Mid-latency Auditory Evoked Potentials

Brain-computer interface technology has been widely studied all over the world,which can help the paralyzed or disabled patients to rebuild the motion control and the ability of communication with the external world.At present,The technology of brain computer interface system based on visual evoked electroencephalogram?EEG?has been widely applied.However,for some patients with visual impairment,the visual stimulation is limited,so,it is very had for them to use the brain computer interface system Fortunately,the auditory system of most patients with visual impairment is usually intact,therefore,it is of potential value to study the brain computer interface system based on auditory evoked electroencephalogram.However,the auditory BCI systems are in the initial stage and there are many defects in the existing system.Therefore,designing a new paradigm of auditory brain computer interface is meaningful.As we all know,there are some problems in the design of the auditory attention paradigm.For example,using a large number of electrodes,needing longer time to evoke P3 potentials and so on.Therefore,an experimental paradigm based on the middle latency response?MLR?evoked from auditory is designed.At first,the corresponding middle latency waveform is induced respectively in two states?attention or non attention?.Then the data is preprocessed by six layers using db10 wavelet transform.Due to the filtered data may still have some interferences,such as electromyographical interference,ocular artifact and so on.So this paper also uses coherent averaging algorithm to filter data again.Finally,it is proved that the method is effective by the comparison of experiments.Since the data of preprocessed auditory evoked potential is still a high-dimensional vector,the feature extraction algorithm is used to reduce the dimension of the collected data.In this paper,based on the autoregressive model?AR?,the AR model coefficients are extracted by the Burg algorithm.And the energy,area,variance,AR model coefficient and waveform peak value of MLRs are respectively calculated.It provides the basis for the using of pattern recognition classification algorithms.Finally,we use support vector machines?SVM?and artificial neural network?ANN?classification algorithms to classify offline data and count the results.The classification accuracy of 8 subjects was counted as well.The experimental results show that in terms of the combined feature₁v,₂v and₃v,the classification accuracy based on ANN is 10.7%,8.7%and11.1%higher than that of SVM,respectively.Of all the results,the optimal classification accuracy can be gained from the combination feature₁v and artificial neural network,which is up to 84.7%.Based on the above results,it can be concluded that the experimental paradigm designed in this paper is effective,and it is rather effective for most of the subjects.This lays a solid foundation for applying this experimental paradigm to BCI in the future.