請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/43794
標題: | 多受試者腦電磁時空獨立成分群聚之研究 Clustering Based Group-level Electromagnetic Spatiotemporal Independent Component Analysis |
作者: | Chia-Hao Chang 張家豪 |
指導教授: | 蔡志鑫(Arthur C. Tsai) |
關鍵字: | 腦電波,腦部訊號源定位,獨立成分分析,電磁時空獨立成分分析,群聚分析, EEG,Inverse problem,ICA,EMSICA,clustering, |
出版年 : | 2009 |
學位: | 碩士 |
摘要: | 腦波紀錄儀 (electroencephalography, EEG) 可用來記錄被激化的神經元在頭皮上所造成的電位差。獨立成分分析 (independent components analysis, ICA) 常被用來分析腦波的訊號,並用以去除原始訊號所包含的雜訊或是其他非大腦皮質活動所產生的訊號;其獨立成分也可被用來估算大腦皮質的活動分佈。籍由對獨立成分的群聚分析 (clustering analysis),我們可從多受試者中找出其相對應的獨立成分,以及其在大腦皮質上相同的活動模式,藉以研究大腦的動態特性。近來,一個新的獨立成分分析方法被提出,稱為「電磁時空獨立成分分析」 (electromagnetic spatiotemporal ICA, EMSICA);其分析方法能直接估計大腦皮質的活動分佈。
在本論文中,我們使用電磁時空獨立成分分析來找出實驗中 9 個受試者腦波訊號的獨立成分,並進一步利用其估算的大腦皮質活動分佈來找出相似的獨立成分。我們也使用傳統的獨立成分分析,並將與獨立成分相應的大腦皮質活動分佈由一個電偶極子來估算,再由電偶極子作群聚分析以作比較。實驗結果顯示,直接使用電磁時空獨立成分分析確實比單個或是數個電偶極子更能正確的估算其大腦皮質的活動分佈,也因此更能正確的將獨立成分分類。最後我們也就如何增進群聚準確度提出一些方法和建議。 Electroencephalography (EEG) is a common technique for recording the electrical activity on the scalp generated by the activation of neurons within the brain. Independent components analysis (ICA) is widely used for eliminating noise and non-brain artifacts by decomposing EEG data into several independent components. Those components can also be used for estimating the activation of neurons. By clustering components across subjects, the common patterns of activations can be identified, which are useful for studying brain dynamics. Recently, a new variant of ICA, called Electromagnetic Spatiotemporal ICA (EMSICA), is proposed, which estimates spatiotemporal independent components and the activation of neurons simultaneously. In this thesis, EMSICA is applied for decomposing EEG data recorded from 9 participants in the experiment, and then the components are clustered according to the sources distributed on the whole cortical surface to find common patterns of activation. Traditional ICA is also applied to the same EEG data for comparison. In ICA, the source configuration for each independent component is represented by one or several equivalent dipoles. The results show that the source distribution estimated directly from EMSICA gives better estimation than equivalent dipoles. We also make some suggestions for improving the accuracy of clustering. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/43794 |
全文授權: | 有償授權 |
顯示於系所單位: | 資訊工程學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-98-1.pdf 目前未授權公開取用 | 11.37 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。