自組織映射圖方法於禪定與休息腦波之Alpha腦分佈圖的分類

SOM Clustering of Alpha Brain Mappings of Zen-Meditation and Resting EEG

李金龍(Jing-Long Lee)

指導教授 ： 羅佩禎

國立交通大學/資訊學院/生醫工程研究所/碩士(2018年)

摘要

本論文中，我們試圖探索禪定和休息的狀態下腦電波(30通道)的α波隨著時間演變的空間特性，本研究採用兩種不同的非監督式分類方法,第一種方法是自組織映射圖 (SOM)，基於輸入特徵向量(α腦映射)和表示輸出神經元的定量特徵來匹配,SOM由30個輸入神經元(對於α腦映射擷取的30通道)和給定數量的輸出神經元建構.因此,我們需要確定提供適當分群的輸出神經元數量,另外,為了優化分群的結果,有必要仔細的選擇參數,例如訓練次數,學習率,鄰近範圍.根據分群的結果,可以確定α腦空間屬性的特徵.模糊C均值(FCM)是一種基於K均值(K-means)的模糊分類器,FCM演算法與k-means的不同之處在於K-means是單一標準的實現.換句話說,為了取代0/1或是真/假,這樣單一的決定,FCM允許每個資料點都有隸屬度是屬於0到1之間.最後,我們比較自組織映射圖(SOM)與模糊C均值(FCM)的分類效能,顯然SOM的群集內距離是73.53對上FCM的79.33,提供較好的群集內聚集效果,群集間的區分是(168.72對上143.32)SOM也具有較高的群集間區分效果.但是,在錯誤分群率卻是FCM的0%更優於SOM的0.09%

關鍵字

腦電波 ； 自組織映射圖 ； 模糊C均值 ； Alpha波

並列摘要

This thesis is aimed to investigate the temporal evolution of alpha spatial properties of 30-channel Zen-meditation and resting EEG (electroencephalograph). Two different schemes of unsupervised classification methods are adopted in this study. The first scheme, SOM (self-organization map) is based on the matching of input feature vector (alpha brain mapping) and the weight vectors representing the quantitative features of the output neurons. The SOM is constructed by 30 input neurons (for the 30 entries extracted from alpha brain mapping) and a given number of output neurons. Accordingly, we need to determine the number of output neurons that provide appropriate clustering. In addition, to optimize the clustering result, it is necessary to carefully select the implementation parameters such as number of training steps and learning rate. From the clustering results, the features of alpha spatial property may be determined. Fuzzy c-means (FCM) is a fuzzy classifier based on the K-means. FCM algorithm differs from the K-means in the aspect that K-means is implemented with the rigid criteria. In other words, instead of reaching a crispy decision like “0/1” or “true/false”, fuzzy scheme allows the degree of truth of a statement to be between 0 and 1. Finally, we compare the performance of the clustering result between SOM and FCM. Apparently, SOM provides the clustering performance of better cohesion (inner bonding, 73.53 against 79.33) and mutual-cluster differentiation (168.72 against 143.32), yet, under the tradeoff of slight falsely-clustered rate (0.09%).

並列關鍵字

Electroencephalograph (EEG) ； Self-organization map (SOM) ； Fuzzy c-means (FCM) ； Alpha rhythm

參考文獻

Reference

[1] S. van Leeuwen, W. Singer and L. Melloni, “Meditation Increases the Depth of Information Processing and Improves the Allocation of Attention in Space,” Frontiers in Human Neuroscience, vol. 6, 2012.

[2] S. Chung, M. Brooks, M. Rai, J. Balk and S. Rai, “Effect of Sahaja Yoga Meditation on Quality of Life, Anxiety, and Blood Pressure Control,” The Journal of Alternative and Complementary Medicine, vol. 18, no. 6, pp. 589-596, 2012.

[3] A. Chiesa and A. Serretti, “A systematic review of neurobiological and clinical features of mindfulness meditations,” European Psychiatry, vol. 25, p. 1044, 2010.

[4] F. Travis and J. Shear, “Focused attention, open monitoring and automatic self-transcending: Categories to organize meditations from Vedic, Buddhist and Chinese traditions,” Consciousness and Cognition, vol. 19, no. 4, pp. 1110-1118, 2010.