Adaptive EEG artifact suppression using Gaussian mixture modeling

Neural tracking using electroencephalography (EEG) recordings suffers from physiologic and extraphysiologic artifacts. We propose an integrated method to adaptively track multiple neural sources while reducing the effects of artifacts. Time-frequency features are first extracted from EEG recordings without pre-processing to suppress artifacts. Unsupervised clustering using Gaussian mixture modeling is then used to separate sources from artifacts, and the clustering results are incorporated into a probability hypothesis density filter to estimate the parameters of an unknown number of sources. Simulation results demonstrate the method's effectiveness in increasing the tracking accuracy performance for multiple neural sources using recordings contaminated by artifacts.