Improved rejection of artifacts from EEG data using high-order statistics and independent component analysis

While it is now generally accepted that independent component analysis (ICA) is a good tool for isolating both artifacts and cognition-related processes in EEG data, there is little definite proof that data preprocessed using ICA is more effective than artifact rejection on raw channel data, especially when more subtle signal processing methods are used to detect artifacts. Here we applied five statistical signal processing methods for detecting artifactual data epochs from either the raw data containing simulated artifacts or from the ICA decomposition of these data, and tested their performance for different sizes of introduced artifacts. The most efficient rejection method used threshold limits applied to the single trial data spectra. We show that for this or other methods ICA preprocessing can improve the detection of data epochs containing eye, muscle, and electrical artifacts by 10-20%.