Kmeans-ICA based Automatic Method for EOG Denoising in Multichannel EEG Recordings

Elie Bou Assi and Sandy Rihana


Neural Sensory Systems and Rehabilitation, Brain Computer Interface, Independent Component Analysis, Kmeans clustering, Electroencephalography


Electroencephalogram (EEG) recordings are contaminated by different internal and external noises and interferences. Therefore, they should be manipulated in order to restore them from these artifacts that could be eye blinks, electrocardiogram (ECG) and many others. Recent research is mainly oriented toward implementing methods in order to remove ocular artifacts whose frequency band overlap with the EEG frequency of interest. Independent Component Analysis (ICA) has already shown to be an effective way for removing the activity of these artifacts. However, when implementing an ICA-based method, the key relies on how to identify the ocular artifact components. Based on the components characteristics, different features such as correlation coefficients, distribution ratio, and maximum value have been identified in order to recognize in an automatic way the artifactual components and their subtraction from the original space to get ocular artifacts free EEG signals. Artifactual components were identified using an adaptive thresholding by means of Kmeans clustering. Qualitative and quantitative techniques of evaluation are presented and give promising results. The classification accuracy based on the correlation feature reached 99.54%.

Important Links:

Go Back