以靜息態腦磁波訊號進行經痛程度的客觀評估

Abstract

疼痛是一種既複雜又主觀的經驗。為了瞭解疼痛的機制及發展疼痛治療的方 法，如何客觀地評估疼痛程度是一個重要的議題。關於疼痛的研究大多著重在外 在刺激所引起的短暫疼痛，但在大多數的日常情況下，疼痛往往是內源且長時間 存在的。所以本研究的目標是利用靜息態的腦磁波訊號發展評估內源疼痛程度的 方法。而此篇論文中，我們將致力於原發性痛經所造成的經痛，所以我們和台北 榮民總醫院合作，取得十四名原發性痛經（是一種不具有骨盆異常的經痛）病患 在月經期休息狀態下的腦波測量資料。 在本研究中我們將使用三種類型的特徵擷取方法。第一種為頻域上的特徵值， 包含中位頻率、頻域邊界頻率、 Shannon 頻域熵、Tsallis 頻域熵和Rényi 頻域 熵。第二類為時序上複雜度的分析，包括Lempel-Ziv 複雜度和多尺度熵。而第 三種則是利用前兩類型的特徵以取得左右半腦非對稱性作為特徵。最佳的特徵為 在前頂葉皮層內側區域所獲得尺度為七十七的多尺度熵，大體上它能用來預測自 我評量的疼痛程度（簡單線性回歸分析，p<0.0007）。之後我們將透過前進選擇 法結合其他的特徵進入多重線性回歸以獲得更好的評估模型。在此我們所選取的 特徵總共有八個，而主要是在前頂葉皮層內側區域較高尺度的多尺度熵。藉由 leave-one-out 交叉驗證法，所提出的方法應用在疼痛程度（由0 至10 的實數 值）的評估可達到0.2 殘差。接著將疼痛程度量化至整數值時，在疼痛評估的表 現上可達到100%的準確率。未來我們希望此篇研究所提出的方法可成為在臨床 上評估內源性疼痛的可靠指標。

Pain is a complex and subjective experience. Therefore, objective assessment of pain scale is essential to the understanding the mechanism of pain as well as the development of pain treatments. There were numerous studies focusing on the induced pain with short duration caused by external stimulus. However, pain is endogenous and lasting in most situations in daily lives. The purpose of this work is to develop an objective metrics for measuring endogenous pain scale by analyzing resting magnetoencephalographic (MEG) signal data. We concentrate on the pain caused by primary dysmenorrhea (PDM, menstrual pain without pelvic abnormality). The data were collected by Taipei Veterans General Hospital from 14 PDM patients at the menstrual phase.

Three categories of features were extracted from the acquired MEG data. The first category was spectral analysis, including relative band powers, median frequency (MF), 90% spectral edge frequency (SEF90), and Shannon (SSE),Tsallis (TSE), and Rényi (RSE) spectral entropies. The second category was temporal complexity analysis, including Lempel-Ziv complexity (LZC) and multi-scale entropies (MSE). The third category of features was hemispheric asymmetric indices of the features in the first and second categories. The best feature, asymmetric index of MSE in temporal scale 77 in anterior medial parietal (amP) region, can be used to adequately predict the self-rating pain scale (simple regression analysis, p=0.0007). Then we combined other features by forward selection. In this step, we mainly incorporated the asymmetric indices of higher temporal scales in amP region. These features were used in the assessment model constructed by multiple linear regression. Through leave-one-out cross-validation, the proposed method achieved 0.2 residual error in predicting pain scale (real value from 0 to 10). After the pain scale value was quantized to integers, it can achieve 100% accuracy in pain assessment by using eight features. We expect that the proposed method could be a reliable indicator for the assessment of endogenous pain scale in clinical applications.