標題: 以靜息態腦磁波所估算得之腦部活動進行經痛程度的客觀評估
Objective Assessment of Menstrual Pain Scale by Resting-state Brain Activity Estimated from MEG
作者: 陳怡地
Chen,Yi-Ti
陳永昇
Chen, Yong-Sheng
資訊科學與工程研究所
關鍵字: 腦磁波;疼痛評估;經痛;休息狀態;樣本熵;楔前葉;額上回;預設模式網絡;多元線性回歸;MEG;Pain assessment;Menstrual pain;Resting state;Sample entropy;Precuneus;superior frontal gyrus;Default mode network;Multiple linear regression
公開日期: 2015
摘要: 客觀的疼痛評估方法對於了解疼痛感產生的機制與發展治療疼痛的方法是 很重要的,但由於疼痛是一種主觀且複雜的體驗,因此要精確且客觀地量化所感 受到的疼痛程度是困難的。大部分關於疼痛的研究皆專注於外在刺激所引發的疼 痛,但其實在日常生活中,很多時候疼痛是由體內產生的。本篇研究嘗試透過分 析靜息態的腦磁波訊號來發展客觀評估內源性疼痛程度的方法,針對原發性痛經 所引發的內源性疼痛,我們使用在台北榮民總醫院對二十五名原發性痛經患者在 月經期間所量測得之休息狀態腦磁波來進行評估,以檢驗所提出方法的可行性。 我們利用最大對比光束構成法估算出大腦皮質的活動,並從大腦皮質的活動 萃取出包含頻域與時序方面的特徵值。頻域方面的特徵值包含中位頻率、頻域邊 界頻率與頻域熵,而時序方面的特徵值包含Lempel-Ziv複雜度和多尺度樣本熵, 我們也會計算上述特徵值在左右半腦的非對稱性指數作為一類特徵值。根據這些 特徵值與我們利用循序向前搜尋法與多變項線性迴歸來建造出疼痛程度評估模 型。我們將資料進行leave-one-out 的分割來交叉驗證疼痛程度評估模型的正確 性。根據我們的實驗,這個評估模型所評估的疼痛程度與病患自評的疼痛程度之 相關係數為0.62 (p = 0.00097),平均殘差為1.39。根據隨機排列檢定,我們 所提出的疼痛程度評估模型之可信度是顯著的 (p = 0.026)。除了討論評估模型 的有效性外,我們也探討了時常被選為評估模型所用的特徵值。我們發現大約百 分之八十的評估模型含有左楔前葉的時序尺度九十五樣本熵與額上回的時序尺 度五十三樣本熵的非對稱性指數,且左楔前葉與額上回皆曾被過去研究提及與疼 痛反應相關。我們希望此篇研究所得到的結果可以在臨床上作為評估內源性疼痛 的可靠指標。
Objective assessment of pain intensity is essential to investigate the mechanism and treatment of pain. However, pain is a subjective and complicated experience. It is challenging to quantify the intensity of perceived pain objectively and accurately. Most of the pain-related studies concentrate on pain induced by external stimulation but not on endogenous pain. In this work, we develop a method to objectively assess the subjective intensity of endogenous pain by analyzing the resting state magnetoencephalographic (MEG) recordings. The MEG data acquired at Taipei Veterans General Hospital for 25 primary dysmenorrhea patients during their menstrual phase were used to evaluate the accuracy of the proposed method. We estimated the cortical source activity from resting MEG recordings by using maximum contrast beamformer and extracted spectral and temporal features from the cortical source activity. Spectral features include relative band power, spectral entropy, median frequency, and spectral edge frequency. Temporal features include multiscale sample entropies (SampEn) and Lempel-Ziv complexity. The hemispheric asymmetric indices of features mentioned above were also calculated as a type of feature. After feature extraction, we applied the sequential forward search and multiple linear regression to construct the pain assessment models. According to leave-one-out cross-validation the correlation coefficient between the predicted pain scores using the pain assessment model and self-rating pain scores was 0.62 (p=0.00097), and the mean residual error was 1.39. The reliability of the proposed pain assessment model was significant in a random permutation test (p=0.026). We also examined the features frequently selected. We found that about 80% of assessment models contained both the SampEn in temporal scale 95 in the left precuneus and the asymmetric index of the SampEn in temporal scale 53 in the medial part of superior frontal gyrus. Both precuneus and superior frontal gyrus have been reported to be associated with pain-related activation in previous studies. The proposed pain assessment model could be used as a reliable indicator for the assessment of endogenous pain intensity in clinical applications.
URI: http://140.113.39.130/cdrfb3/record/nctu/#GT070156063
http://hdl.handle.net/11536/125658
Appears in Collections:Thesis