Single-trial-EEG-based BCI using ICA, ECDL and DBN

点击翻译按钮获取中文摘要

A new single-trial-EEG-based Brain-Computer Interface (BCI) was developed using independent component analysis (ICA), equivalent current dipole source localization (ECDL) and dynamic Bayesian Network (DBN). Quantification Method of type II (QM2) and the BN were used as learning model for categorical data. All the QM2 models for the seven subjects had correlation ratios of more than 0.9, yielding the good discrimination between left and right hands to be imagined, and the correction rates in LOOCV of more than 0.8. Using the obtained BN, the classification rule was established on the basis of the conditional probabilities in the "motor cortex" nodes. Moreover, the generalization to the DBN allowed us to apply to the Bereitschaftpotential, and led to the better performance of the classification rule.