Prediction of Obstructive Sleep Apnea Events from Snoring Using Convolutional Neural Networks

• Main Authors: Sheng-Yen Chen, 陳聖言
• Other Authors: Jen-Chun Lin
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/38j6r2

碩士 === 元智大學 === 電機工程學系甲組 === 107 === Obstructive sleep apnea (OSA) is a sleep disorder, which causes a complete or partial airway obstruction due to pharyngeal collapse during sleep. It is closely associated with a number of serious illnesses, including neuropsychiatric disorder, arterial hypertension, cardio¬vascular disease, stroke, and metabolic syndrome. As the snoring sound is an essential signal of OSA, this study proposes an OSA detection method based on the audio signal using the designed Two-channel Convolutional Neural Network. Specifically, we proposed a two-channel convolutional neural network architecture with 82-layer based on the spectrogram and the Difference-based MFCCs features (TSD model) to recognize apnea events. Due to such two-channel designs, the TSD can therefore effectively discriminate between the snoring and apnea events. The dataset was recorded overnight from 21 patients and the polysomnography results were used to label the snoring signal. On the testing dataset, the classifier achieved a sensitivity and specificity of 82.8% and 81.0%, respectively. Our results indicate that using such a method could help to detect apnea events effectively and show the feasibility for clinical applications. Keywords—Neural Networks, Convolutional Neural Network (CNN), Spectrogram, Mel-Frequency Cepstral Coefficients(MFCC), Snore Sounds, Polysomnography(PSG), Obstructive Sleep Apnea(OSA)