Patterns Extraction Of Mental EEG And Evoked Potentials Based On Independent Component Analysis

The purpose of EEG signal processing is to extract the hidden or weak patterns that probably have some physiological and/or psycho-physiological significance from EEG signals in sophisticated noise background and then to apply them to the research on clinical medicine or cognitive science.The traditional approaches for EEG signal processing are mostly on the time or frequency analysis. But due to the strong randomness and nonstationarity of EEG, the results obtained from those traditional methods are not very satisfying. Independent Component Analysis (ICA) is a multi-dimensional statistical analysis method developed in the 90's of the 20th century and used to analysis the mutually independent nongaussian signals. When some certain assumptions are satisfied, ICA can effectively separate the independent source signals from the synchronous multichannel recording, and then according to some independent sources, we can further analyze their physiological significance. Therefore, as a new promising approach of blind source separation (BSS), ICA has attracted extensive attention of researchers in the international field of signal processing.In this paper, we made an in-depth study on the basic theory, algorithm and application of ICA with the intention of artifacts removing and patterns extraction on EEG and ERP. The innovated works we have finished are as follows:1. Study the Infomax algorithm, extended Infomax and online Infomax algorithm of BSS based on the thought of information transmission maximum theory. And then compare the performances of the above three algorithms by the Matlab simulation results.2. ICA algorithm is used to the artifacts removing and patterns extraction of mental EEG signals from different mental tasks. The results show that ICA can effectively detect, separate and remove a wide variety of artifacts from EEG recordings, such as EOG, ECG By studying the EEG independent sources and their projections on human scalp, we can find that some steady independent components always appear when the subject repeats the same mental tasks.3. ICA is also applied to the physiological patterns extraction of Evoked potentials (EPs). Many studies employ EP peak measures to test clinical or developmental hypotheses. However, EPs cannot be easily decomposed into functionally distinct components since their time courses and scalp projections generally overlap. Our experiment results imply that ICA can be used to effectively decompose and extract multiple overlapping components from sets of related EPs.