Rules-Based and SVM-Q Methods With Multitapers and Convolution for Sleep EEG Stages Classification

Abstract

Sleep EEG signals analysis is an approach that helps researchers identify and understand the different phenomena concealed within sleep EEG data. This research introduces a time-frequency analysis approach to untangle the parameters of the sleep stages classification from EEG data. This approach computes the spectral estimation of a signal based on a set of controlled wavelets using a multitaper with convolution (MT&C) method. In this study, the MT&C methods is implemented to extract the features from a single sleep EEG data channel. Then two separated approaches are applied for sleep stage classification. The first one is based on the EEG waves characteristic definitions of sleep stages (named as Rules-based method) to directly classify each 30 second EEG segment after the feature extraction. The second approach uses a support vector machine with quadratic equation (SVM-Q) classifier to classify the sleep stages based on experts' scoring. The experimental results are evaluated, and the outcomes show an overall accuracy of 90% with an average sensitivity of 96.2% and an average specificity of 93.2% using an SVM-Q classifier and an 87.6% accuracy for the Rules-based method on healthy subjects. On the other hand, the accuracy on subjects with abnormal sleep EEG data is of 78.1% with the SVM-Q classifier and 73.4% with the Rules-based method.

Automatic sleep spindles identification and classification with multitapers and convolution

“…This research implements the SAMC method to identify and classify spindles. The SAMC method uses the MT&C method [ 5 ] to extract the key spindles information from different channels from the three sleep EEG DBs. The proposed method identifies the signal power on frequencies between 11 and 16 Hz.…”

“…The SDE accentuates the signal time–frequency characteristics based on the parameters of the tapers. The tapers are wavelets generated using a Gabor kernel, which is convoluted with the signal, intending to highlight the spectra density of the spindles in sleep EEG data [ 5 ]. The tapers in the MT&C are generated using the Gabor function (1) and the parameters of the spindles [ 44–46 ].…”

“…Multitapers (MTs) are mechanisms of exploration that use time–frequency analysis to extract detailed information from signals and map specific elements of an object or concern. As implemented in [ 5 ], MTs were used to identify different frequencies, power, and time of an event present in sleep EEG data to generate features that were directly associated with the sleep physiology based on the R&K rules [ 25 ]. They classified sleep stages with an average rate of 87% with the option of visualizing them using a spectral estimation from each epoch.…”

“…This paper analyses sleep spindles and their identification from sleep electroencephalogram (EEG) data. The proposed method is a time–frequency analysis using a multitapers and convolution (MT&C) method [ 5–7 ] to calculate the spectral estimation of spindles’ characteristics in EEG data. Sleep spindles are bursts of energy ranging between 9 and 16 hertz (Hz) frequencies with a pyramidal-like structure that wanes and waxes its oscillations between 0.5 and 2 seconds (s).…”

A Detailed Analysis & Parametric Comparison Of Eeg Processing Models

“…Researchers may more precisely identify and grasp sleep EEG events by analyzing sleep EEG signals [39]. The objective of this work is to use time-frequency analysis to classify EEG sleep stages.…”