Different neural networks approaches for identification of obstructive sleep apnea

Abstract

Obstructive sleep apnea (OSA) is one of the most common breathing-related sleep disorders affecting individuals of different age groups, genders and origins. It is characterized by short-duration of cessations in breathing during sleep due to the collapse of the upper airway. The golden standard and reliable test for the detection of OSA is conducted by specialized physicians performing a polysomnographic sleep study. However, this test is time/labor consuming, expensive and cumbersome. In this paper, a non-invasive technique employing three different artificial neural networks to analyze spectral and statistical features of the Heart Rate Variability (HRV) signal to identify OSA subjects from normal control is investigated. The artificial networks include the single perceptron network, the feedforward network with back-propagation and the probabilistic neural network. The highest performance on MIT standard data is achieved by the feedforward network with back propagation using wavelet-based frequency domain features with specificity, sensitivity, and accuracy of 90%, 100% and 96.67%, respectively.