Diagnosis of sleep apnea by optimal fuzzy system based on respiratory signals

This Sleep apnea is an important and common disorder that leads to the reduction or interruption of respiratory flow during sleep due to obstruction of the airways. Currently, nocturnal polysomnography (PSG) is the standard approach for diagnosing sleep apnea. This method is expensive, time-consuming and inconvenient due to space limitations and the complexity of recording physiological signals. Today, machine learning algorithms are used to provide more friendly and accessible solutions that help in cost-effective and easy to diagnosis. In this article, a fuzzy network optimized with artificial intelligence algorithms is formed as a classband model for sleep apnea diagnosis. Changes in the acoustic signals caused by breathing sounds during sleep apnea can be used to monitor and diagnose sleep apnea events. The dataset of audio signals recorded from the sound of breathing is collected by a microphone and a mobile phone. The fuzzy classband model was built based on Mamdani's fuzzy inference rules from the extracted features, and the correct classification rate (CCR) was 77.5 by adjusting different parameters of the model. Coefficients were obtained as feature values to increase data resolution by the particle swarm optimization (PSO) optimization algorithm and Fisher's discriminant ratio. By applying these coefficients to the input of the fuzzy network in order to optimize the feature vector, the correct classification rate was obtained equal to 87.5% with the leave one out cross-validation method.