Sliding window higher-order cumulants for detection of eye blink artifact from short segments of single-channel EEG

Background The prevalence of single-channel electroencephalogram (EEG) is increasing. However, eye blink artifacts can contaminate EEG signals, potentially impacting their clinical utility. Method This study presents a sliding window higher-order cumulant (HOC) method for accurate blink artifact detection. The method operates in two steps. First, preliminary detection of blink artifacts is achieved by computing HOCs within the sliding windows of contaminated EEG. Second, the method identifies blink peaks within the detected intervals and further adjusts the detection results using the peaks. Results The study compares the effectiveness of the sliding window HOC method with the existing variational mode extraction (VME) and multi-window summation of derivatives within a window (MSDW) methods using semi-simulated and real data. The comparison results show that, for semi-simulated data, the sliding window HOC method exhibits the highest Youden index values in terms of the direct accuracy metrics. When it comes to the indirect metrics of artifact reduction effectiveness based on the three detection methods, the sliding window HOC method has the highest correlation coefficient (CC) (0.935 vs . 0.914, 0.909), the lowest relative root mean square error (RRMSE) (0.128 vs . 0.186, 0.225), and the lowest mean absolute error (MAE) (1.634 vs . 1.770, 1.748). For real data, it yields the highest MAE ratio of the lower-to-higher frequency bands among three datasets. The algorithm also demonstrates computational efficiency with millisecond-level resolution. Discussion The sliding window HOC method provides superior detection performance and greater robustness. The presented method provides a solid foundation for effectively removing blink artifacts in single-channel EEG.