Discriminative methods based on sparse representations of pulse oximetry signals for sleep apnea–hypopnea detection. (March 2017)

• Record Type: Journal Article
• Authors: Rolón, R.E.; Larrateguy, L.D.; Di Persia, L.E.; Spies, R.D.; Rufiner, H.L.
• Is Part Of: Biomedical signal processing and control. Volume 33(2017)
• Page Start: 358
• Page End: 367
• Publication Date: 2017-03
• Subjects: Sleep apnea–hypopnea syndrome -- Sparse representations -- Dictionary learning -- Neural networks; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2016.12.013
• Languages: English
• ISSNs: 1746-8094

Abstract

Abstract : Graphical abstract: Abstract : Highlights: Two novel methods to detect apnea–hypopnea events are presented. Sparse representations of only pulse oximetry signals are used. Discriminative information of sparse coefficients is successfully extracted. A neural network for detecting apnea–hypopnea events is used. Abstract: The obstructive sleep apnea–hypopnea (OSAH) syndrome is a very common and generally undiagnosed sleep disorder. It is caused by repeated events of partial or total obstruction of the upper airway while sleeping. This work introduces two novel approaches called most dicriminative activation selection (MDAS) and most discriminative column selection (MDCS) for the detection of apnea–hypopnea events using only pulse oximetry signals. These approaches use discriminative information of sparse representations of the signals to detect apnea–hypopnea events. Complete (CD) and overcomplete (OD) dictionaries, and three different strategies (FULL sparse representation, MDAS, and MDCS), are considered. Thus, six methods (FULL-OD, MDAS-OD, MDCS-OD, FULL-CD, MDAS-CD, and MDCS-CD) emerge. It is shown that MDCS-OD outperforms all the others methods. A receiver operating characteristic (ROC) curve analysis of this method shows an area under the curve of 0.937 and diagnostic sensitivity and specificity percentages of 85.65 and 85.92, respectively. This shows that sparse representation of pulse oximetry signals is a very valuable tool for estimating apnea–hypopnea indices. The implementation of the MDCS-OD method could be embedded into the oximeter so as to be used by primary attention clinical physicians in the search and detection of patients suspected of suffering from OSAH.