以腦電訊號預測腦部損傷病患之預後與通道選擇機制

Abstract

缺氧缺血型腦病變與中風是因為腦部血流供應受阻，氧氣與養分供給不足所造成的腦部損傷疾病。若能早期預測預後或分析嚴重度，將能幫助醫護人員提供病患適當且有效的治療。腦電圖是在人的頭皮上以非侵入的方式貼電極，量測腦部神經的放電情形，這項工具已被廣泛的使用在反映大腦的功能。然而要解讀這項生理訊號需仰賴專業護理人員的經驗與觀察，因此提供量化腦電圖指標來進行後續分析是一明確且需要的目標。但若將所有電極所蒐集到的生理訊號都做分析，不帶有損傷資訊的訊號可能會影響我們後續分析的分類正確性。此外，冗長且複雜的電極擺放前置作業除了造成病患的不適，也可能影響醫護人員在使用這項分析工具上的意願。 本論文中，我們引入一非線性方法-多尺度熵進入腦電訊號分析處理流程中。流程包含去除訊號干擾以及背景雜訊去除，可以從受雜訊干擾的原始生理訊號還原出乾淨的訊號，再從中萃取出損傷資訊。此外，為了避免多餘的通道訊號進入分析，我們根據經由傳統影像診斷工具如電腦斷層掃瞄或是核磁共振，由專業醫護人員所標記的損傷位置，提出通道選擇技術來解決此問題。此分析架構應用於加護病房中的缺氧缺血型腦病變與中風病患，其接收者操作特徵曲線下面積皆能達到0.8，意味著本論文提出的腦電訊號分析架構有潛力適用於預測缺氧缺血型腦病變之預後與急性中風病患之嚴重程度。

Hypoxic-ischemic encephalopathy (HIE) and stroke are syndromes of brain injury resulting from critical reduction or loss of blood flow and supply of oxygen and nutrients. Early outcome prediction or progressive severity monitoring can help medical staff to offer more adequate and effective treatment. Electroencephalography (EEG) activity represents cerebral function theoretically. However, interpretation of EEG signals is limited to visual inspection by medical staff. Therefore, there is a clear and necessary need to provide quantitative EEG (qEEG) parameters for analysis. Furthermore, applying all EEG channels may takes the irrelevant or redundant channels containing no lesion information into the following analysis which may degrade the classification accuracy. Prolonged and complex preparation can cause discomfort to the patient and impact the medical stuff’s willingness to use EEG devices. In this thesis, we introduce a nonlinear method – multiscale entropy (MSE) into our proposed EEG signal processing flow which consists of signal de-noising and de-trend. These pre-processes help us to sift clean signals from noises comprehensively so that we can extract the information of brain injury in the following analysis. Then, to avoid redundant information involved, we propose a channel selection method into the EEG analysis flow according to spatial information of lesion’s location. By using the EEG-based framework to analyze the HIE and stroke patients’ EEG signals in intensive care unit (ICU), we find that the area under receiver operating characteristic (ROC) curves > 0.8, implying that the proposed analysis framework has potential for predicting outcome of HIE and severity of stroke.