| Sleep spindles are distinct oscillations observed in specific frequency bands,often 11Hz-16 Hz,and in specific sleep stage named none rapid eye movements sleep.Traditionally,sleep spindles detection usually relies on subjective visual inspection of electroencephalogram(EEG)signals by experts,but this task is very time consuming and is not a robust enough approach for intra-and intersubject variability.With the increasing interest for sleep spindles detection and the increasing amount of EEG recording data,the need for automatic sleep spindle detectors increases.Compared with the visual detection,the automatic sleep spindle detectors are faster,more reproducible,and systematic.However,there some issues with automated sleep spindle detectors also emerge,showing their weak robustness when spindles characteristics change and the weak agreement with gold standard given by sleep experts.The aim of this work is to analyse the fashionable automatic spindle detectors,through examples of the performance of nine published automated sleep spindle detectors.The results of the automated detectors show great intersubject variation and fragile performance under different test metic.Since automatic sleep spindle detectors are mainly characterized by their signal decomposition and decision making processes,a preprocessing step is usually necessary to improve the performance of the sleep spindle detection algorithm.The dual-basis pursuit denoising technique was applied in this paper to improve the spindle detection methods,reducing the inconsistency of the performance induced by individual variations and making up the short board for the automated methods in some specific testing metic.In addition to testing the preprocessing,this paper combines the multi-objective evolutionary algorithm and the dual-basis pursuit denoising method to find the optimal solution space of the automated sleep spindle detection algorithms.The electroencephalogram data of eight subjects in the DREAMS database were used to test the performance of algorithm and the results of the two expert evaluation in the database were used as the gold standard,and our algorithm was evaluated using eight test scores.The method using dual basis pursuit denoising preprocessing raw EEG signal and multiobjective evolutionary algorithm finding the optimal values of parameters of automatic spindle detectors is easy to imply and useful in the polysomnography and sleep health detector systems. |
