Abstract

Electroencephalography (EEG) technique has been widely used in continuous monitoring the brain activities in academic research and clinical applications. In cognitive neuroscience research, the electrical brain signals can be used to measure mental effort of subjects. However, the presence of artifacts is a constant problem when recording brain activities, which will obscure the underlying neural dynamics and therefore make it difficult to interpret EEG signals accurately. These unwanted signals or artifacts have different effects depending on their sources of generation. Among them, the motion of the subject is one of the major contributors to physiological artifacts that causes most of the contaminations to the underlying brain activities. It is quite challenging to correct the myogenic activity from EEG background potentials due to its wide spectral distribution overlapped with typical bands of brain waves related to cognitive activities, and the spatial distribution over the entire scalp of human. As such, this thesis focuses on the analysis and removal of motion artifacts from EEG signals.
The preliminary investigations include the movement-triggered artifact identification and the analysis of the characteristics of the motion artifact. According to the recorded video, the contaminated epochs are extracted from the original EEG signals. A set of features of the movement-triggered artifacts are proposed based on power spectral density and wavelet transform. Statistical analysis is performed to distinguish the segments that contain motions. Two typical methods of artifact removal are then studied, and the efficiency to correct this type of artifact is validated by comparing the extracted features of non-movement segments and the contaminated segments. The result shows that the tested artifact removal methods cannot completely remove movement artifacts, which also infers the potential relation between motion and mental activities.