Brain activities are often investigated through Electroencephalographic (EEG) data analysis using time-domain Independent Component Analysis (ICA). Nevertheless, the instantaneous mixing model of ICA cannot properly describe spatio-temporal dynamics, such as those related to traveling waves of neural activity. In this work, we exploit the application of the Complex ICA (cICA) to describe the underlying brain activities in time and frequency domain. In particular, we show how to effectively extract the most significant time-frequency structure of cortical activity in order to solve a compelling EEG-based pattern classification problem. The crucial step of independent component selection among frequencies is performed using an objective computational method based on template matching techniques with physiologically-plausible activations. Experimental results are obtained using on-line EEG data from the BCI Competition 2003 and are expressed in terms of confusion matrix after leave-one-out validation procedure. A comparative analysis between ICA and cICA models reveals that cICA estimation gives powerful information and allows to achieve a higher classification accuracy with respect to instantaneous ICA.
Decoding underlying brain activities in time and frequency domains through complex independent component analysis of EEG signals
VALENZA, GAETANO;VANELLO, NICOLA;SCILINGO, ENZO PASQUALE;LANDINI, LUIGI
2014-01-01
Abstract
Brain activities are often investigated through Electroencephalographic (EEG) data analysis using time-domain Independent Component Analysis (ICA). Nevertheless, the instantaneous mixing model of ICA cannot properly describe spatio-temporal dynamics, such as those related to traveling waves of neural activity. In this work, we exploit the application of the Complex ICA (cICA) to describe the underlying brain activities in time and frequency domain. In particular, we show how to effectively extract the most significant time-frequency structure of cortical activity in order to solve a compelling EEG-based pattern classification problem. The crucial step of independent component selection among frequencies is performed using an objective computational method based on template matching techniques with physiologically-plausible activations. Experimental results are obtained using on-line EEG data from the BCI Competition 2003 and are expressed in terms of confusion matrix after leave-one-out validation procedure. A comparative analysis between ICA and cICA models reveals that cICA estimation gives powerful information and allows to achieve a higher classification accuracy with respect to instantaneous ICA.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
