Detection of obstructive sleep apnoea using dynamic filter-banked features

Abstract

There is a need for developing simple signal processing algorithms for less costly, reliable and noninvasive Obstructive Sleep Apnoea (OSA) diagnosing. One of the promising directions is to provide the OSA analysis based on the heart rate variability (HRV), which clearly shows a non-stationary behavior. So, a feature extraction approach, being capable of capturing the dynamic heart rate information and suitable for OSA detection, remains an open issue. Grounded on discriminating capability of frequency bands of HRV activity between normal and OSA patients, features can be extracted. However, some HRV normal spectrograms resemble like pathological ones, and vice versa; so, prior to extract the feature set, the energy spatial contribution contained in each subŨ band should be clarified. This paper presents a methodology for OSA detection based on a set of short-time feature banked features that are extracted from the spectrogram of the HRV time series. The methodology introduces the spectral splitting scheme, which searches for spectral components with alike stochastic behavior improving the OSA detection accuracy. Two different splitting approaches are considered (heuristic and relevance-based); both of them performing minute-by-minute classification comparable with other outcomes that are reported in literature, but avoiding more complex methods or more computed features. For validation purposes, the methodology is tested on 1-min HRV-segments estimated from 50 Physionet database recordings. Using a parallel combining k-nn classifier, the assessed dynamic feature set reaches as much as 80% value of accuracy, for both considered approaches of spectral splitting. Attained results can be oriented in research focused on finding alternative methods used for less costly and noninvasive OSA diagnosing with the additional benefit of easier clinical interpretation of HRV-derived parameters.

Extraction of Stationary Spectral Components Using Stochastic Variability

“…Several variability analysis techniques suitable for clinical applications had been proposed, including statistical, geometric, energetic, informational, and invariant measures [2]. In this research, the amount of stochastic variability of the spectral component set is computed following the approach given in [4], that is based on time-variant decomposition estimated by adapting in time any of commonly used latent variable techniques, upon which a piecewise stationary restriction is imposed [6]. So, under the locally stationary assumption, consistent estimates of the time-varying spectral density matrix are obtained and consequently consistent estimates of the time-varying eigenvalues and eigenvectors may be accomplished [5].…”

“…The database holds a available collection of 1-min HRV segments selected from Physionet, which holds 70 electrocardiographic recordings, each one including a set of reference annotations added every minute of the recording indicating either the presence of absence of apnoea during each segment. Finally, 600 HRV segments of 1-minute length (300 apneic and 300 normal labeled) were selected from 25 training recordings to build the dataset [4].…”

“…In biosignal applications, it is often of interest to be able to separate an observed time series into two or more groups with different stochastic behavior [4]. In particular, there is a need for distinguishing stationary from non-stationary components, either because its assumption is a pre-requisite for applying most of standard algorithms devoted to steady-state regimes, or because its breakdown conveys specific information in evolutive contexts, as remarked in [8,1].…”

“…Commonly, the time domain non-stationariness can be quantitative measured by the variation of either first two statistical moments over time (second order process). Instead, the non-stationary upon the spectrum domain of biological signals has been emphasized more recently, because of the corresponding physiological meanings of some spectral components [3], for instance, for Heat Rate Variability (HRV) recordings, the spectral components can provide information about the relative inputs of the two autonomic nervous system mutually antagonistic components the sympathetic one, which acts to increase heart rate, and the parasympathetic component acting to slow the heart and to dilate blood vessels [4]. Nevertheless, even the traditional time-averaged spectrum of nonstationary signals, being widely used, is not appropriate for quantifying the temporal variability in the spectrum.…”

Finite Rank Series Modeling for Discrimination of Non-stationary Signals

“…Based on generalized spectral representation, a more elaborated approximation, termed Finite-rank Series Modeling, can be carried out by decomposing the underlying biosignal into a linear combination of products of harmonics, exponential, or even polynomial series [2], that is, the problem is addressed as modeling multiple time-evolving data streams governed by some linear recurrent formula, and inferring similarities and relationships between these data over prescribed time window lags. Although, a single input signal can be decomposed by an unknown number of recurrent sequences, which may vary from observation to observation, a strong constrain on this decomposition approach is related with modeling time series holding itself different stochastic structures [3].…”