Fast STF model and applications on EEG analysis

Abstract

Searching for the tool that can efficiently summarize a multi-channel EEG signal is a challenging problem in EEG processing. In this paper, we propose the fast implementation of the 3-way parallel factor analysis (PARAFAC) called Fast STF model (fSTF model) which can simultaneously employ all the space, time, and frequency domains of a multi-channel EEG. The multi-channel EEG signal is first subdivided along space and time domains into the selected numbers of segments. By carefully selecting the number of segments according to the structure of the brain, signatures (features) extracted from the fSTF model are comparable with those from the conventional STF model while the time used in computation is reduced by more than 50%. Signatures obtained from the fSTF model are further summarized as a single number to indicate the quality of the multi-channel EEG signal. The simulation results illustrate the merits of the proposed model via the applications on eyeblink artifact-contaminated EEG decomposition and EEG quality assessment.