Sleep spindles detection using Empirical Mode Decomposition

Sleep spindles are very important EEG patterns in modern neuroscience. There were developed many spindle detection algorithms, but not all of them are suitable for patients with insomnia because of artifacts, movements and complicated spindle producing. The paper presents a spindle detection method based on proper preprocessing and classification of stationary segments using Naive Bayes classifier. Preprocessing was performed using Empirical Mode Decomposition, which decomposes the signal into trends. Trends rejecting from the signal gives filtered signal for feature processing. To evaluate the quality of proposed approach, F-measure, positive predicative value and true positive rating were calculated. The method shows good results on dataset of 11 insomniac patient: F-measure by sample was 40.72% and F-measure by events was 48.59%. The results were also compared with Martin, Molle, Wendt and Ferallelli methods.