Multiscale AM-FM methods on EEG signals for motor task classification

In this manuscript, we present the use of customized, multiscale amplitude-modulation frequency-modulation (AMFM) methods on electroencephalography (EEG) brain signals during the subject development a motor task: right hand and left hand. This approach is compared to various non-linear patterns and methods that have been applied in order to characterize and understand the dynamic behavior of the EEG signals. The AM-FM methods have been optimized in terms of multiscale filters for the mu band (8-12 Hz). The instantaneous AM-FM values are processed using their probability density function and classified using multiple layer perceptron (MLP) and the partial least squares regression (PLS). The system is tested using the standard BCI dataset with results with a precision to 89% and an area under the ROC to 91%.