Electroencephalogram (EEG) signals are obtained from the scalp surface to study various neuro-physiological functions of brain. Often, these signals are obscured by the other physiological signals of the subject from heart, eye and facial muscles. Hence, the successive applications of EEG are adversely affected. The wide spectrum and high amplitude variation of muscle artifact overlaps EEG both in spectral and temporal domain.

In this paper, an adaptive singular spectrum analysis (SSA) algorithm is proposed to remove muscle artifact from single channel EEG. The mobility threshold for the SSA routine is decided adaptively using a neural network regressor (NNR). The NNR is trained using the features of contaminated EEG with various levels of contamination for better approximation of the reconstructed EEG signal.

The proposed algorithm is validated using both simulated and experimental data. Parameters like relative root mean square error ( ), correlation coefficient ( ), peak signal to noise ratio ( ), and mutual information (MI) along with graphical results are used to evaluate the performance of the proposed algorithm. The proposed algorithm is found to be having consistent and better performance while the other algorithms show a decline in performance with high level of contamination.

The algorithm upon testing with both simulated and experimental data, is able to discriminate between various contamination levels present in EEG and performed comparatively better than the existing single channel algorithms.