A robust motifs based artifacts removal technique from EEG

Abstract

Artifact removal is one of the important pre-processing tasks in electroencephalogram (EEG) signal analysis. Some of the artifacts generated by muscular and ocular movements are present at specific EEG segments. The magnitudes of these artifacts are too high to retrieve any information underneath. Hence, they need to be completely removed before any meaningful processing of the signals. The existing methods do not address this issue; rather, use correlated methods, which corrupt the other segments along with the information content of the EEG signals. In this paper, a combination of temporal motifs and dynamic time warping (DTW) method is used to identify specific types of artifact segments quickly. A comparison between variants of DTW algorithms such as derivative DTW (DDTW), weighted DTW (WDTW), weighted DDTW and fast DTW (FDTW) has been carried out for identifying repetitive artifacts. The identified artifacts are removed, and the missing samples are reconstructed without affecting the other segments of the EEG. The FDTW has found to have minimum execution time for finding the accurate location of the repetitive artifacts. The proposed motif based artifacts removal technique yields to result in fast and precise removal of artifacts that can be used both in the clinical and research perspective of EEG analysis. Motifs based artifacts removal technique along with traditional methods such as least mean square, recursive least square and fast independent component analysis are also implemented for the sake of comparison. The average time consumption and efficacy measure in terms of signal-to-noise ratio in dB scale have been found out. The DTW based algorithms found to have outperformed the traditional methods.