Independent Component Analysis On Brain Computer Interface

As a new interactive communication and control system, brain computer interface attracts a widespread attention and discussion because of its application on medical rehabilitation area. Independent component analysis, a novel signal processing method, promotes the development and progress of BCI technology.This article emphasizes on applications of the independent component analysis in brain computer interface, especially in source localization problems. After a brief introduction of ICA and source localization theory, it focuses on the use of ICA in pre-processing, feature extraction and source inverse problems. In addition, real SSVEP data is analyzed and processed with the help of EEGLAB.In this paper, the data firstly is preprocessed by re-reference, removing baseline and filtering so as to reduce noise preliminarily. Moreover, Event-related potential is extracted to strengthen the evoked information and remove noise further. Due to the multi-dimensional property, Principle Component Analysis is applied to reduce the dimensionality so as to maintain useful information fully. In the next part, different ICA algorithms are performed such as Fixed-point and Extended-Infomax to get independent components.Based on a priori knowledge, various artifacts such as EoG or ECG are identified and removed initially. Besides that, Linkage and K-Means are implemented on independent components respectively to complete the clustering of them. With the combination of clustering results and subjective experience, artifacts are removed clearly.New data is obtained after artifacts removing. And then brain computer source analysis is done in DIPFIT according to different head models. It proves that the generating position of visual stimulation assembled in hindbrain area and verifies the accuracy and reliability of ICA method on source localization.