Spectrogram classification of patient chin electromyography based on deep learning: A novel method for accurate diagnosis obstructive sleep apnea

Abstract

Obstructive sleep apnea (OSA) is the most common sleep-related breathing disorder and the cause is an increased upper airway resistance during sleep, leading to partial or complete interruption of airflow. It is important to find and treat OSA because it is caused starting of many dangerous illnesses. For diagnosing OSA, polysomnography (PSG) is the gold standard but it is complex, costly, and time-consuming Hence, a simple and automated recognition system can be very useful. Numerous studies have been conducted on sleep-disordered breathing (SDB) based on a variety of signals and algorithms in recent years. Researchers have always attempted to intelligently diagnose OSA with fewer and more accessible signals using faster and simpler algorithms due to the complexity of biological signals and the difficulty of visual recognition and interpretation. In this paper, a novel and effective technic is proposed by using artificial intelligence. This study introduced and examined for the first time the idea of applying chin electromyogram signal directly and independently to diagnose OSA. Two-dimensional spectrograms generated from the chin electromyogram of 100 patients were fed as input to three pre-trained deep learning models and their performances were compared with a multilayer perceptron network. Results showed that our proposed newfound approach outperforms the current states and it proved chin electromyography spectral analysis and classification of produced 2D spectrograms with the deep learning model provides an effective, rapid, and accurate diagnosis and prediction tool in this field with an accuracy of more than 99%.