摘要
In order to explore the correlation between the adjacent segments of a long term EEG, an improved principal component analysis(PCA) method based on mutual information algorithm is proposed. A one-dimension EEG time series is divided equally into many segments, so that each segment can be regarded as an independent variables and multi-segmented EEG can be expressed as a data matrix. Then, we substitute mutual information matrix for covariance matrix in PCA and conduct the relevance analysis of segmented EEG. The experimental results show that the contribution rate of first principal component(FPC) of segmented EEG is more larger than others, which can effectively reflect the difference of epileptic EEG and normal EEG with the change of segment number. In addition, the evolution of FPC conduce to identify the time-segment locations of abnormal dynamic processes of brain activities,these conclusions are helpful for the clinical analysis of EEG.
In order to explore the correlation between the adjacent segments of a long term EEG, an improved principal component analysis （PCA） method based on mutual information algorithm is proposed. A one-dimension EEG time series is divided equally into many segments, so that each segment can be regarded as an independent variables and multi-segmented EEG can be expressed as a data matrix. Then, we substitute mutual information matrix for covariance matrix in PCA and conduct the relevance analysis of segmented EEG. The experimental results show that the contribution rate of first principal component （FPC） of segmented EEG is more larger than others, which can effectively reflect the difference of epileptic EEG and normal EEG with the change of segment number. In addition, the evolution of FPC conduce to identify the time-segment locations of abnormal dynamic processes of brain activities, these conclusions are helpful for the clinical analysis of EEG.
基金
Natural Science Foundatoin of Fujian Province of China
grant number:2010J01210,2012J01280
