Single-Channel EEG Signal Enhancement in Presence of EMG artifact using ELM-based Regressor

Abstract

Electroencephalogram (EEG) used to read the electrical signals from human scalp for diagnostic purposes. The EEG electrodes sensitive, so the low amplitude EEG signals get corrupted by the wide spectrum and high amplitude electromyogram (EMG) signals. Hence, the recorded EEG have segments that have artifacts with the unacceptable state. Effectively recovering the corrupted signal from a single channel EEG is a challenge. The proposed algorithm enhances the single-channel EEG signal in the presence of EMG artifacts using extreme learning machine (ELM) regressor. For training and testing of the ELM network, EEG signals are subjected to S-transform and the obtained transformation matrix is used as the feature set. S-Transform has the advantage of uniquely combining the gradual resolution and complete referenced phase information for the subjected time series. The ELM is trained using both magnitude and phase of corrupted and clean EEG signals in pairs. This training can reduce the EMG artifact from corrupted EEG signals effectively and enhance the same in the testing stage. The evaluation parameters used for the proposed algorithm are the average root mean square error (RMSE) and the correlation coefficient (CC) between the ground truth EEG signal to the estimated EEG signal. The average RMSE and CC were found to be 0.260 and 0.97 respectively for the simulated dataset.