Evaluation and extraction of mismatch negativity through exploiting temporal, spectral, time-frequency, and spatial features
Fengyu Cong · Jyväskylä University Digital Archive (University of Jyväskylä) · 2010
This study is dedicated to the evaluation and extraction of mismatch negativity (MMN) from EEG recordings through exploitation of its temporal, spectral, time-frequency and spatial features. Compared to the conventional wavelet transformation, Hilbert-Huang transformation can more effectively represent the temporal and spectral features of MMN. To qualify the correlation of recordings among single trials (CoRaST), the proposed frequency domain MMN model with Fourier transformation only requires less than one percent of the computational time which the conventional inter-trial coherence costs, promising to implement CoRaST in real-time systems. In the time domain, a robust and reliable method to extract the MMN component is proposed through exploitation of its temporal, frequency and spatial information. The core of this novel method includes the wavelet decomposition (WLD) and independent component analysis (ICA). Thus, it is denoted as wICA. The proposed wICA can outperform the conventional difference wave using the temporal information, optimal digital filter using the frequency band information of MMN and ICA using the spatial information. In the time-frequency domain, nonnegative matrix factorization (NMF) and nonnegative tensor factorization (NTF) can decompose the time-frequency represented BEG to obtain the desired time-frequency represented MMN and P3a components simultaneously, releasing the independence assumption among sources which ICA requires, and discovering more knowledge between normal children and clinical children in our study. Moreover, the proposed methods in this study may also be usable in the research into other MMNs and other event-related potentials since they may also have similar temporal, spectral, time-frequency and spatial features as the MMN attended to in this study.