以單導程腦電圖訊號進行睡眠覺醒事件偵測之研究

Abstract

過去幾年來智慧醫療領域越來越受重視，許多資訊科技包括雲端、遠距、大數據及人工智慧等，都已經大量應用於醫療領域。應用在病患監測服務、行動照護的需求大增，而睡眠診斷這一塊在其中受到非常大的重視，睡眠影響人的生活作息，要是沒有好的睡眠，人體就會失去正常機能運作。 在本國，睡眠醫療專家在診斷睡眠覺醒以及睡眠呼吸中止症這些症狀的方式還是必須讓病人在醫院睡一個晚上，利用睡眠多項生理儀(PSG)紀錄病人在這一個晚上的腦波圖、肌電圖、心電圖、眼動圖、血氧飽和濃度、胸腹呼吸動作還有口鼻呼吸氣流這些諸多的生理特徵，接著再由兩至三位睡眠專家根據以上的生理紀錄進行討論並且判斷這一位病人在某些時段是屬於正常的狀況或是發病的狀態，但這個階段非常曠日費時，甚至不同的睡眠專家之間也會有不同的意見，因此本論文透過近年快速崛起的深度學習以及機器學習技術，提出一套系統化、並且可靠、客觀的方式，在觀察大量資料之下，利用單一導程的腦電圖訊號，成功的將病人的睡眠覺醒(Sleep Arousal)狀態分辨出來，提供睡眠醫療專家參考，輔助監測病人身體狀態。 本論文首章節為論文簡介，第二章以及第三章為背景知識介紹，從第四章開始為本論文的主要貢獻，也就是系統架構設計，包含腦波訊號的前處理、特徵抽取、分類器架構…等，並在第五章探討模擬測試結果，末章則為結語與未來展望。

In the past few years, the field of smart health has been more and more important. Lots of technology such as cloud computing、remote control、big data and artificial intelligence has already applied to medical diagnosis. Particularly, sleep problems are a frequent complaint among many people, especially the elderly, and have a substantial impact on quality of their lives. Sleep arousal conventionally refers to a temporary intrusion of wakefulness into sleep or at least a sudden transient elevation of the vigilance level due to arousal stimuli. And, many researches have shown that sleep arousals can induce various sleep disorders. Thus, arousals are a good marker of sleep disruption representing a harmful feature for sleep. Nowadays, the method of detecting sleep arousals is to collect patient’s physiological data such as electroencephalography、electrocardiography and electromyography through an overnight polysomnography test. After getting these recordings, two to three sleep experts would score the arousal regions according to their background knowledge. However, manual scoring of arousals conducted by sleep experts is generally time-consuming and subjective. Moreover, different experts are very likely to have different opinions. Therefore, the objective of this study is to develop an algorithm for automatic detection of sleep arousals based on single lead electroencephalography (EEG) by using machine learning and deep learning methods.