Factors affecting discrepancies between scorers in manual sleep spindle detections in single-channel electroencephalography in young adult males

Here, we aimed to clarify the factors that cause individual differences in manual spindle detection during sleep by comparing it with automatic detection and to show the limitations of manual detection. Polysomnography (PSG) signals were recorded from ten young male participants, and sleep stages were classified based on these signals. Using time-frequency analysis, we detected sleep spindles from the single-channel electroencephalography (EEG) of C4-A1 within the same PSG data. Our results show a detailed accuracy evaluation by comparing the two skilled scorers' outputs of automatic and manual sleep spindle detection and differences between the number of sleep spindle detections and spindle time length. Additionally, based on automatic detection, the distribution of Cohen's kappa for each scorer quantitatively showed that individual scorers had detection thresholds based on EEG amplitude. Conventionally, automatic detection has been validated using manual detection outputs as the criterion. However, using automatic detection as the standard and analyzing the manual detection outputs, we quantitatively showcased the differences in individual scorers. Therefore, our method offers a quantitative approach to examining factors contributing to discrepancies in sleep spindle detection. However, individual differences cannot be avoided when using manual detection, and automatic detection is preferable when analyzing data to a certain standard.