A robust two-stage sleep spindle detection approach using single-channel EEG

摘要

Objective. Sleep spindles in the electroencephalogram (EEG) are significant in sleep analysis relatedto cognitive functions and neurological diseases, and thus are of great clinical interests. Anautomatic sleep spindle detection algorithm could help decrease the workload of visual inspectionby sleep clinicians. Approach. We propose a robust two-stage approach for sleep spindle detectionusing single-channel EEG. In the pre-detection stage, a stable number of sleep spindle candidatesare discovered using the Teager energy operator with adaptive parameters, where the number oftrue sleep spindles are ensured as many as possible to maximize the detection sensitivity. In therefinement stage, representative features are designed and a bagging classifier is exploited to furtherrecognize the true spindles from all candidates, in order to remove the false detection in the firststage. Main results. Using the union of all experts’ annotations as the ground truth, its performanceoutperforms state-of-the-art works in terms of F1-score (F1) on two public databases (F1: 0.814 forMontreal archive of sleep studies dataset and 0.690 for DREAMS dataset). The annotationconsistency between the proposed method and certain selected expert as the trainer could exceedthe consistency between two human experts. Significance. The proposed sleep spindle detectionmethod is based on single-channel EEG thus introduces as less interference to the subjects aspossible. It is robust to subject variations between databases and is capable of learning certainannotation rules, which is expected to help facilitate the manual labeling of certain experts. Inaddition, this method is fast enough for real-time applications.