A Multivariate Empirical Mode Decomposition based Filtering for Subject Independent BCI

Goal: A brain-computer interface (BCI) provides a way to translate the motion intentions of human using brain signals such as electroencephalogram (EEG) into control commands. EEG signals are highly subject specific and ... [ view full abstract ]

Goal: A brain-computer interface (BCI) provides a way to translate the motion intentions of human using brain signals such as electroencephalogram (EEG) into control commands. EEG signals are highly subject specific and non-stationary. One of the most challenging tasks is to classify motion intentions since the recorded EEG signals have inherent non-stationarities which are due to changes in the signal properties over time within as well as across sessions. Thus it becomes difficult to achieve a stable operation of BCI. Method: We present a novel filtering method based on the multivariate empirical mode decomposition (MEMD) using subject independent BCI (MEMD-SI-BCI) for classification of motor imagery (MI) based EEG signals to achieve enhanced BCI. A subject independent BCI can be used immediately by the new user without using the user’s training data. The MEMD method helps to utilize the cross channel information and enhance localization properties. It decomposes multichannel EEG signals into a set of multivariate intrinsic mode
functions (MIMFs). These MIMFs can be considered narrowband, amplitude and frequency modulated (AM-FM) signals. The statistical property, namely, mean frequency measure of these MIMFs has been used to combine these MIMFs to compute the enhanced EEG signals which have major contributions due to
µ and β rhythms over the motor cortex region. The objective of the proposed method is to filter EEG signals before feature extraction and classification to enhance the features separability and ultimately the BCI task classification performance. The common spatial pattern (CSP) feature has been computed from the enhanced EEG signals and has been used as a feature set for classification of left hand and right hand MIs using a linear discriminant analysis (LDA) based classification method. Results: We have achieved an improvement of >11% in the evaluation stage using the MEMD-SI-BCI method when compared with SI-BCI. Significance: This study helps to develop BCI systems with intuitive motor imaginations, thus facilitates broad use of noninvasive BCIs. We have evaluated our method on publicly available BCI competition IV dataset 2A and have obtained improved performance.