Automated sleep staging of OSAs based on ICA preprocessing and consolidation of temporal correlations

Abstract

An automated sleep staging based on analyzing long-range time correlations in EEG is proposed. These correlations, indicating time-scale invariant property or self-similarity at different time scales, are known to be salient dynamical characteristics of stage succession for a sleeping brain even when the subject suffers a destructive disorder such as Obstructive Sleep Apnea (OSA). The goal is to extract a set of complementary features from cerebral sources mapped onto the scalp electrodes or from a number of denoised EEG channels. For this purpose, source localization/extraction and noise reduction approaches based on Independent Component Analysis were used prior to correlation analysis. Feature extracted segments were then classified in one of the five classes including WAKE, STAGE1, STAGE2, SWS and REM via an ensemble neuro-fuzzy classifier. Some techniques were employed to improve the classifier’s performance including Scaled Conjugate Gradient Method to speed up learning the ANFIS classifiers, a pruning algorithm to eliminate irrelevant fuzzy rules and the 10-fold cross-validation technique to train and test the system more efficiently. The performance of classification for two strategies including (1) feature extraction from effective cerebral sources and (2) feature extraction from selected channels of denoised EEG signals was compared and contrasted in terms of training errors and test accuracies. The first and second strategies achieved 92.23 and 88.74% agreement with human expert respectively which indicates the effectiveness of the staging system based on cerebral sources of activity. Our results further indicate that the misclassification rates were almost below 10%. The proposed automated sleep staging system is reliable due to the fact that it is based on the underlying dynamics of sleep staging which is less likely to be affected by sleep fragmentations occurred repeatedly in OSA.

Appendix

The algorithm of WTMM method to find the singularity spectrum consists of the following steps:

• A set of lines of local modulus maxima of the wavelet coefficients, L(a), is found at each scale a.

• The partition functions are calculated by the sum of the power of order q of the modulus maxima of the wavelet coefficients along each line at the scales smaller than the given value a,

  $$Z(q,a)=\sum\limits_{{l \in L({a^*})}} {{{\left( {{{\sup }_{a \leq a}}|W({a^*},{b_l}({a^*})|} \right)}^q}}$$

  (11)

  where ${b_l}({a^*})$ determines the position of the maximum corresponding to the line l at each scale.

• As $a \to 0$ (small scales), the partition function shows the power low behavior, $Z(q,a)={a^{\tau (q)}}$. Now, the scaling exponent $\tau (q)$ can be extracted as the slope of a log–log plot of the partition function versus the scale a; ${{\tau (q) = \log _{{10}} Z(q,a)} \mathord{\left/ {\vphantom {{\tau (q) = \log _{{10}} Z(q,a)} {\log _{{10}} a}}} \right. \kern-\nulldelimiterspace} {\log _{{10}} a}}$.

• A regression is then adopted to estimate $\tau (q)$ as a linear function of q for monofractal signals,$\tau (q)=qh(q) - 1$, where $h(q)=d\tau (q)/d(q)=const$ is the global Hurst exponent. For multifractals,$\tau (q)$ is a nonlinear function, $\tau (q)=qh(q) - D(h)$, with a large number of Holder exponents describing local scaling of the wavelet coefficients.

• The singularity spectrum (distribution of local exponents) can be expressed using a Legendre transform $D(h)=qh(q) - \tau (q)$ which quantifies the statistical properties of the different subsets characterized by different exponents [30]. An example of local maxima lines determination by WTMM method for a 30s sleep EEG segment is show in Fig. 6.

Fig. 6

Example of local maxima lines determination by WTTM for a 30-s sleep segment