Selection of Nonstationary Dynamic Features for Obstructive Sleep Apnoea Detection in Children

Abstract

This paper discusses the methodology for selecting a set of relevant nonstationary features to increase the specificity of the obstructive sleep apnea detector. Dynamic features are extracted from time-evolving spectral representation of photoplethysmography envelope recordings. In this regard, a time-evolving version of the standard linear multivariate decomposition is discussed to perform stochastic dimensionality reduction. For training aim, this work analyzes the concrete set comprising filter banked dynamic features that include spectral centroids, the cepstral coefficients as well as their timevariant energies. Performance of classifier accuracy is provided for the collected polysomnography recordings of 21 children. Moreover, since the apnea diagnosing is based on analysis of set of fragments partitioned from the photoplethysmography envelope recordings, a new approach for their indirect labeling is described. As a result, performed outcomes of accuracy bring enough evidence that if using a subset of cepstral-based dynamic features, then patient classification accuracy can reach as much as 83.3% value, when using a k-nn classifier, as well. Therefore, photoplethysmography-based detection provides an adequate scheme for obstructive sleep apnea diagnosis.

Selection of time-variant features for earthquake classification at the Nevado-del-Ruiz volcano

“…So, the following time-variant relevance measure is assessed, as discussed in Sepulveda-Cano et al (2011):…”

“…However, the set of measured time-variant features can be large containing a considerable amount of redundancy, which in turn leads to either overtraining or a significant increase in the computational burden. In such a situation, dimension reduction must be strongly considered to select the adequate number of relevant features either by encoding or removing both the redundant and irrelevant information (Sepulveda-Cano et al, 2011). Á lvarez et al (2012 applies a discriminative feature selection method on different sets of parameters for classification of volcano-seismic signals to select the minimum number of features containing most of the discriminative information.…”

Training using short-time features for OSA discrimination

“…From the database, 25 recordings were used as a training set for the classification algorithms. A second group with 25 recordings was used as a test set to measure the performance of the algorithms, as considered in [9].…”

“…Parameter tuning for a considered TFR is achieved by the procedure developed for biosignals, discussed in [9]. the STFT-based quadratic spectrogram is computed by sliding Hamming windows for the following set of estimation parameters: 32.5 ms processing window length, 50% of overlapping, and 512 frequency bins.…”

Finite Rank Series Modeling for Discrimination of Non-stationary Signals

“…Mostly, a methodology for analysis of non-stationary time series is based on the assumption that there is a processing time window of such a length that the piecewise stationarybased approach of analysis holds. Although determination of proper stationary data length and the model parameters remains as an open issue [1]. Grounded on piecewise stationary approach, several time-variant linear decomposition techniques had been proposed for non-stationarity characterization: time-frequency representation, smoothing techniques based on ortogonal polynomials, linear splines, wavelets, empirical mode decomposition, and regressive modeling, among others.…”