Independent component analysis for EOG artifacts minimization of EEG signals using kutosis as a threshold

• Main Author: Islam, Kazi Aminul
• Format: Dissertation
• Language: English
• Published: ProQuest Dissertations & Theses 01-01-2016
• Subjects: Electrical engineering; Medical imaging
• ISBN: 9781369079111; 1369079117

Brain electrical activity commonly represented by the Electroencephalogram (EEG), due to its miniscule amplitude (on the order of a hundred microvolts), is often contaminated with various artifacts. Independent Component Analysis (ICA) may be a useful technique to minimize the artifacts prior analyzing the original neural signal. Independent Component Analysis method has been used to separate the recorded EEG signal to original neuronal signal and artifact affected noisy signal. These independent components have been investigated further to separate and identify to find the noisy components among them. We used kurtosis to determine the threshold to separate the artifacts-affected ICA components from the unaffected components. Kurtosis may represent how peaked or how flat the artifacts that affect a signal are compared to the normal behavior of the original signal. To select the threshold value of the kurtosis, two statistical principles have been used: namely, the Z-score and the confidence interval. Our intention was to avoid a manual technique to determine the affected ICA components and, instead, to explore an automatic method based on the kurtosis value. Based on the observed results, we may conclude that the present technique may be used for EOG artifacts minimization. Kurtosis values, median values, max and min values of different channels have been used to train neural network. After much iteration when the artificial neural network has been trained to desired accuracy, additional EEG signals have been used to test the network outputs. From the output result we may conclude that trained neural network predicted noisy channels accurately.