Multi-dimensional PARAFAC2 component analysis of multi-channel EEG data including temporal tracking

The identification of signal components in electroencephalographic (EEG) data originating from neural activities is a long standing problem in neuroscience. This area has regained new attention due to the possibilities of multi-dimensional signal processing. In this work we analyze measured visual-evoked potentials on the basis of the time-varying spectrum for each channel. Recently, parallel factor (PARAFAC) analysis has been used to identify the signal components in the space-time-frequency domain. However, the PARAFAC decomposition is not able to cope with components appearing time-shifted over the different channels. Furthermore, it is not possible to track PARAFAC components over time. In this contribution we derive how to overcome these problems by using the PARAFAC2 model, which renders it an attractive approach for processing EEG data with highly dynamic (moving) sources.