Efficient Blinking Component Estimation in Subspace-Based EEG and MEG Analysis

Removing EyeBlink (EB) artifact is often an essential preprocessing step of EEG/MEG analysis. Subspace-based methods such as Independent Component Analysis (ICA) and Signal Source Projection (SSP) are well-known methods of EB artifact removal. However, these methods require preknowledge of the number of components (NoC) involved in the blinking segments which is unknown and estimated by visual inspection. Here, we provide a reliable and efficient method for estimating the NoC of EB. The method utilizes an SVD denoising approach denoted by Number of Source Eigenvalue Error (NoSEE). The proposed method performance is examined by synthetic EEG, and is validated by real MEG data. Root Mean Square Error (RMSE) is used to evaluate the performance while visual inspection on real MEG data is used to validate the algorithms results. In both cases, the proposed method provides optimum NoC in the sense of MSE minimization and consistency with the real data. The method provides more accurate results compared to manual methods. In addition, the low computational complexity of the algorithm results in much less processing time for this automatic NoC calculator compared to the trial and error procedure of the manual methods which enables the algorithm to be used in online EEG analysis.