腦電圖頻譜特徵於阿茲海默症中之探討

Abstract

非侵入性的腦電圖（EEG）技術，在阿茲海默症（AD）的診斷上展現出良好的應用潛力。本論文分為A、B兩部分，分別探討alpha波段去同步化與功率頻譜中的1/f成分，兩者在文獻中皆為與認知功能相關的腦電圖特徵。在A部分中，AD組（n = 79，年齡 = 74.1 ± 5.9歲）相較於健康對照組（n = 77，年齡 = 65.7 ± 4.7歲），表現出顯著較低的alpha波段能量以及通道間相干性，並且從閉眼過渡至睜眼的過程中，其alpha去同步化反應也較為減弱。這些發現意味著AD患者在局部和長距離的功能性網路中，皆存在神經同步活動受損的情形。 在B部分中，我們提出一項稱為ZCM（Zero-Crossing Modified FOOOF）的方法，用以將EEG功率頻譜拆解成1/f與震盪兩個成分。ZCM 利用一次微分分析排除震盪波峰，以引導1/f成分的初始擬合。模擬結果顯示，ZCM方法在整體準確度上並未優於原始FOOOF方法。雖然ZCM能降低估計1/f指數參數時的系統性偏差，但此優勢因其對雜訊較為敏感、導致顯著較高的估計變異性所抵銷。然而，在真實腦波資料中，ZCM針對原始方法難以準確描述的頻譜，展現較佳的擬合度。 總結來說，本論文呈現了相干性可以作為alpha波段去同步性的量化指標。而在頻譜參數化方面，針對ZCM的探討提供了關於微分擬合方法限制的見解。結合兩者，這些結果提供了頻譜特徵，有助於提升對AD病理的瞭解。

Electroencephalography (EEG) is a noninvasive tool that has proven promising for the diagnosis of Alzheimer’s disease (AD). Divided into Part A and Part B, this thesis was aimed to investigate alpha desynchronization and the 1/f component of EEG power spectrum, respectively, both of which were previously reported to associate with cognitive functions. In Part A, AD patients (n = 79, age = 74.1 ± 5.9 years) were found to exhibit significantly lower alpha-band energy and inter-channel coherence, along with diminished alpha desynchronization during the transition from eyes-closed to eyes-open states, when compared with normal controls (n = 77, age = 65.7 ± 4.7 years). The findings suggested impaired neural synchronization in both local and long-range functional networks in AD. In Part B, a modified method termed Zero-Crossing Modified FOOOF (ZCM) was proposed to decompose EEG power spectrum into 1/f and oscillatory components. ZCM leveraged first-derivative analysis to exclude oscillatory peaks during the initialization of 1/f fitting. Simulation results indicated that ZCM did not yield superior overall accuracy compared to the original FOOOF method. While the proposed method reduced systematic bias in estimating the 1/f exponent, this advantage was offset by significantly higher estimation variance due to increased sensitivity to noise. However, in real-world EEG data, ZCM demonstrated improved goodness-of-fit specifically for spectra that were poorly characterized by the original method. In summary, this thesis has demonstrated that coherence is useful as a quantitative index of alpha desynchronization. Regarding spectral parameterization, the investigation of ZCM provides insights into the limitations of derivative-based fitting. Together, the results provide spectral signatures that may advance the understanding of AD pathology.