Neural networks for the novel diagnosis method based on clinical snoring sound analysis in apnea.

Obstructive Sleep Apnea (OSA) is the most common type of sleep apnea. OSA is caused by an obstruction in the upper airway, which actually stops the air flow in the nose and mouth during sleep. The current gold standard of diagnosis, called polysomnography (PSG), requires a fullnight hospital stay connected to over ten channels of measurements requiring physical contact with sensors. PSG is inconvenient, expensive and unsuited for community screening. OSA is commonly accompanied by snoring, but its potential in clinical diagnosis is not fully recognized yet. In this paper, we use the snore sounds (SS) as diagnostic systems, and propose the novel technique to track the sleep state of OSA patient in the connection weight space (CWS) of neural networks (NN) after the process of supervised training. While the SS data/database of OSA patient provides an excellent medium to test our methods, we carry out experiments using two variations of the basic back-propagation (BP) algorithm: the steepest descent (SD) algorithm and the Levenberg-Marquardt (LM) algorithm. This paper compares the performance of SD with that of LM algorithm in an attempt to find an ideal artificial NN-training algorithm to model the SS. [ABSTRACT FROM AUTHOR]