Signal Processing Techniques For Artifact Removal In Electroencephalogram (EEG)

The electroencephalographic recordings measure the electric impulses generated in the brain, in response to a given stimulus. The spontaneous EEG data is used for diagnosis and treatment of some brain diseases. For the data to be used for clinical applications, it needs to be free of the various artifacts like the eye blinks, movement, head movements and muscle activity. These artifacts need to be corrected or the affected parts need to be removed in the preprocessing of the EEG dataset. This pre-processing is normally done manually, which tends to be not only time-consuming but also subjective. With large number of datasets to be analyzed, it is necessary to have uniformity in the analysis. Uniformity, reproducibility and reliability in the preprocessed data can be obtained if a statistical approach is taken while preprocessing the datasets. Ideally, this can be semi- or fully- automated. This approach therefore, needs to be taken while removing the less frequently occurring artifacts and correcting the more frequently occurring artifacts, so as to retain more complete datasets for further research or clinical purposes. This thesis covers the entire span of EEG data preprocessing and data quality assurance. It emphasizes the correction of eye blink artifact, one of the most frequently occurring artifacts. The spatial filter method, which makes use of the underlying brain activity data segment while computing its filter coefficients, is introduced as an effective approach for correcting ocular artifacts. This spatial filter described is based on the spatial distribution of the eye blink over the entire scalp region. In order to detect and reject subtle artifact, a novel set of signal attributes are proposed that describe the head movements, horizontal eye movements, and spurious bad electrodes. The resultant data obtained after the pre-processing steps are clean, i.e. free of artifact contamination free. In order to quantify the results, data was visually inspected after each step of EEG data preprocessing. Instances of the artifacts in each step were visually identified before and after preprocessing. The results of the visual inspection done by an expert in EEG data analysis were then validated with the results obtained from the automated preprocessing method developed. The results obtained by manual as well as semi-automated preprocessing method matched perfectly, with the semi-automated method not only taking less time for computations but also increases the reproducibility of the data.