Postersession 2
Poster #: 29
Topic: Computational models
Thursday, Sep 10, 2015
1st floor

Filtering event-related potentials in time, frequency and space domains sequentially and simultaneously

Fengyu Cong1, Tapani Ristaniemi2, & Heikki Lyytinen3

1Department of Biomedical Engineering, Dalian University of Technology, Dalian, China
2Department of Mathematical Information Technology, University of Jyväskylä, Finland
3Department of Psychology, University of Jyväskylä, Finland

Event-related potentials (ERPs) in the context of electroencephalography (EEG) for the cognitive neuroscience are usually produced through averaging single-trials of preprocessed EEG, and then, the interpretation of underlying brain activities is based on the ordinarily averaged EEG. We find that randomly fluctuating activities and artifacts can still present in the averaged EEG data, and that constant brain activities over single trials can overlap with each other in time, frequency and spatial domains. Therefore, before interpretation, it will be beneficial to further separate the averaged EEG into individual brain activities.

We propose the systematic approaches including pre-process wavelet transform (WT), independent component analysis (ICA), and nonnegative tensor factorization (NTF) to filter averaged EEG in time, frequency and space domains to sequentially and simultaneously obtain the pure ERP of interest, particularly for mismatch negativity. Furthermore, WT is the preprocessing for ICA and NTF respectively in the time and in the time-frequency domains. WT-ICA is suitable for the individual-level data processing and filters ERPs in the time, frequency and space domains sequentially. The filtered ERP data by WT-ICA can be used for group-level data analysis. While, WT-NTF is appropriate for the group-level data processing and analysis, and filters ERPs in the time, frequency and space domains simultaneously.