A Novel EEMD-CCA Approach to Removing Muscle Artifacts for Pervasive EEG

Future electroencephalogram (EEG) recordings in body sensor networks are prone to be contaminated by muscle activity due to the mobile, long-term, and pervasive monitoring needs. In this paper, a novel approach for muscle artifact removal in EEG is proposed by combining ensemble empirical mode decomposition (EEMD) with canonical correlation analysis (CCA), termed as EEMD-CCA. This approach can make good use of inter-channel information. We tested the approach on simulated, semi-simulated, and real-life data sets, respectively. The approach outperformed state-of-the-art techniques, including independent component analysis, CCA, and EEMD-ICA. Statistical tests demonstrate the significance ( $p < 0.01$ ) in (semi)-simulated studies. The relative root-mean-squared error can be reduced to around 0.3 and the average correlation coefficient can be kept above 0.9 even when the contamination is quite heavy (SNR < 2). Besides, we also tested the approach on few-channel EEG randomly selected from multichannel EEG, and obtained competitive results. The computational cost satisfies the real-time requirement. This indicates that the proposed EEMD-CCA approach is applicable under both multichannel and few-channel settings. It is an effective and efficient signal processing tool for enhancing the signal of interest in both hospital and home healthcare body sensor networks.