ABSTRACT
OBJECTIVE: Due to its high temporal resolution, electroencephalography (EEG) has become a broadly-used technology for real-time brain monitoring applications such as neurofeedback (NFB) and brain-computer interfaces (BCI). However, since EEG signals are prone to artifacts, denoising is a crucial step that enables adequate subsequent data processing and interpretation. The aim of this study is to compare manual denoising to unsupervised online denoising, which is essential to real-time applications. METHODS: Denoising EEG for real-time applications requires the implementation of unsupervised and online methods. In order to permit genericity, these methods should not rely on electrooculography (EOG) traces nor on temporal/spatial templates of the artifacts. Two blind source separation (BSS) methods are analyzed in this paper with the aim of automatically correcting online eye-blink artifacts: the algorithm for multiple unknown signals extraction (AMUSE) and the approximate joint diagonalization of Fourier cospectra (AJDC). The chosen gold standard is a manual review of the EEG database carried out retrospectively by a human operator. Comparison is carried out using the spectral properties of the continuous EEG and event-related potentials (ERP). RESULTS AND CONCLUSION: The AJDC algorithm addresses limitations observed in AMUSE and outperforms it. No statistical difference is found between the manual and automatic approaches on a database composed of 15 healthy individuals, paving the way for an automated, operator-independent, and real-time eye-blink correction technique.
