Adaptive EOG Artifact Removal Algorithm Using a Spectrum Magnitude Template Matching for Single-Channel EEG

Removing ocular artifacts (OAs) from electroencephalogram (EEG) signals is crucial for EEG analysis and brain-machine interface (BMI) applications. However, current portable EEG devices encounter significant challenges in artifact removal due to limited electrode configurations. In this article, we propose an adaptive EOG artifact removal algorithm that enhances artifact removal through the suppression of contaminated signals from the frequency domain. The algorithm involves constructing a set of alternative spectrum templates extracted from training datasets, which allows for adaptive selection of the most relevant template for new contaminated EEG data based on Euclidean distance. With the most relevant spectral templates, it facilitates precise artifact removal, while significantly improving signal integrity. We have validated our algorithm by comparing it with existing methods on both semi-simulated and real EEG signal datasets, achieving a notable signal-to-noise ratio (SNR) improvement of 7.14 dB and a correlation increase of 0.12. Furthermore, our computational complexity analysis demonstrates a low runtime of 0.34 ms, reinforcing the algorithm’s suitability for real-time artifact removal in practical BMI applications.