[Research on electrooculography artifact removal from electroencephalography signals based on frequency slice wavelet...

To address the strong contamination of single-channel electroencephalography (EEG) signals by electrooculography (EOG) artifacts during acquisition, this paper proposes an EOG artifact suppression method based on an improved frequency slice wavelet transform (FSWT). First, the EEG signal is adaptively decomposed into frequency-slice components using FSWT. Based on the low-frequency energy bursts induced by eye blinks, a robust median absolute deviation (MAD)-based threshold is used to localize artifact-contaminated intervals. Subsequently, only within the detected artifact-contaminated intervals, the dominant low-frequency range of eye-blink artifacts (0.8～3.5 Hz) is used as a prior reference, and adaptive local time-frequency masking is applied to the corresponding frequency components to reduce excessive removal of low-frequency components in normal EEG signals. Finally, the cleaned EEG signal is reconstructed via inverse FSWT, forming a three-stage artifact-removal framework of "detection-masking-reconstruction". Experiments were conducted on the publicly available semi-simulated Klados dataset using signal-to-noise ratio improvement (ΔSNR) and band energy retention rate as evaluation metrics. In addition, the applicability of the proposed method was validated using self-collected single-channel EEG data acquired under real-world conditions. Results showed that the proposed method achieved mean ΔSNR values of (2.81 ± 2.80) dB for all contaminated samples ( n = 265) and (5.69 ± 1.88) dB for strongly contaminated samples ( n = 101) on the Klados dataset. These values were numerically higher than those obtained using three single-channel methods: empirical wavelet transform (EWT), complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), and wavelet thresholding. The theta-band energy retention rate was 0.889, indicating that the method retained a relatively high proportion of theta-band energy while suppressing EOG artifacts. Results from the self-collected single-channel EEG data further showed that the proposed method could reduce the high-amplitude fluctuations caused by eye blinks while preserving the main EEG rhythm structures under real-world acquisition conditions. In summary, the proposed improved FSWT method can be used for EOG artifact suppression in single-channel EEG signals without relying on multichannel spatial redundancy.