Study on classification of sleeping breath sounds and evaluation of breathing quality

寝息呼吸音の分類と呼吸の質の評価に関する研究

Study on classification of sleeping breath sounds and evaluation of breathing quality

Degree Grantors Yamaguchi University

Sleep is an essential physiological process for the human body. People spend about one-third of their lives sleeping. Both sleep duration and sleep quality are important to human health. Sleep quality describes how restful and restorative the sleep process is. Over 80 sleep disorders are known to affect sleep quality. Among them, sleep-related breathing disorder (SRBD) is the second factor. Sleep-related breathing disorders are sleep disorders in which breathing abnormalities occur during sleep. Abnormal snoring and respiratory arrest or abnormally low breathing during sleep reduce oxygen levels in the blood, increasing the risk of depression, cardiovascular disease, stroke and even death. Therefore, monitoring and analysis of respiration during sleep is gaining increasing importance in healthcare.
Polysomnography (PSG) is considered the gold standard for diagnosing sleep disorders, but PSG is usually performed in an unfamiliar sleep laboratory under the supervision of a medical technician and is often worn with many sensors that interfere with sleep. It is often the case. This research group is developing a breathing sound measurement system that constantly monitors the quality of sleep in general home environment. This system can easily measure breath sounds during sleep all night with high accuracy without disturbing sleep. The purpose of this research is to develop a technique to classify patterns of breathing sounds and to analyze the quality of breathing in order to more accurately analyze the state of sleep from breath sound information. There are various patterns of sleep breath sounds, such as normal breath sounds and snoring, and abnormal breath sounds and snoring. To develop a method to classify these patterns, to develop an algorithm to calculate ventilation from breath sounds, to estimate the sleep apnea index (AHI), and to assess the quality of breathing during sleep. try. Specifically, the temporal feature waveform (TCW) is calculated after partly removing the noise of the breathing sounds of sleep with a band-pass filter. Based on the time feature waveform, a respiratory signal effective for analysis is extracted from low-level signals and phase-divided into a respiratory phase and an apnea or low signal. Mel-frequency cepstrum coefficients (MFCC) are then obtained for the respiratory phases, and an agglomerative hierarchical clustering (AHC) algorithm is applied to distinguish between normal/abnormal breathing, normal/abnormal snoring, and normal/abnormal breathing. , tossing and turning, etc., which are less relevant to breathing. The categorized breathing patterns are analyzed every 30 seconds and the relative tidal volume of the breath is calculated. In addition to verifying the effectiveness and accuracy of the technology and analysis method proposed in this study, a method of estimating the apnea syndrome index (AHI) and converting the ventilation volume into high, medium, and low levels, We propose a method to evaluate the quality of breathing in a patient and verify its effectiveness.
This paper consists of six chapters, including an introduction and conclusion.
Chapter 1 introduces the background and overview of this research.
Chapter 2 describes a signal-processing technique for analyzing breath sounds during sleep and a method for classifying breathing patterns. Breathing sound data during sleep often includes disturbed breathing due to bruxism or body movement, ambient environmental noise, etc. In this chapter, the Time Characteristic Waveform (TCW) and the Characteristic Moment Waveform (CMW) are calculated for respiratory sound signals that have undergone preprocessing, such as filtering noise to preprocess the respiratory sounds, and the segmentation of inspiration and expiration is performed. The Mel-Frequency Cepstrum Coefficients (MFCC) are obtained for each respiratory cycle and applied as a feature vector to the Agglomerative Hierarchical Clustering (AHC) algorithm. This method is used to classify ordinary respiratory signals (normal and abnormal breathing, normal and abnormal snoring) from signals less relevant to respiration, such as tossing and turning and environmental noise.
In Chapter 3, using the technology described in Chapter 2, breathing sound data during sleep are classified into apnea, hypopnea, normal breathing, abnormal breathing, normal snoring, and abnormal breathing for each 30-second frame. In addition, we describe a method for classifying events such as no snoring and rolling over and determining the respiratory state.
In Chapter 4, we propose a method for estimating the apnea-hypopnea Apnea-Hypopnea Index (AHI) for classified abnormal breath sounds and low-level breath sound signals, compare it with the diagnostic results of PSG, and examine its validity. And verify usefulness.
Chapter 5 describes a method for estimating ventilation volume from breath sounds. Because normal breath sounds are correlated with ventilation, this study used a quantitative approach to calculate normal breathing and normal snoring and a qualitative method to calculate apnea/hypopnea and abnormal breath sounds. We will propose and compare it with the diagnosis result of PSG and verify its validity.
In Chapter 6, as an application development, an example of applying the breathing sound classification method proposed in this study to heart sound analysis is presented. Finally, we will explain the construction of a data collection distribution system for sharing auscultation data collected at different facilities and hospitals using blockchain technology.
Chapter 7 presents the conclusions and prospects of this study.

Creators Wang Lurui