腦電波與腦磁波訊號已被廣泛地應用於大腦神經科學學術研究與臨床診斷。然而，非侵入式的腦電/磁波資料，可能夾帶不同頻帶的自發性腦波律動，生理雜訊和系統雜訊。如何從混合訊號中萃取出感興趣之腦波特徵是相當重要的課題。本論文主旨為開發以經驗模態分解法為基礎之腦電/磁波分析技術，達到單一試驗分析腦電/磁波的目的。經驗模態分解法根據資料本身的特質，自適性的拆解出不同階層之訊號。根據各個應用設計出事件相關成分辨識流程，以降低非事件相關成分之干擾。本研究中，利用經驗模態分解法分別應用於視覺腦波大腦人機界面、大腦感覺運動皮質區mu波萃取與嗅覺誘發電位萃取。在頻率特徵萃取的應用中，本研究利用穩態視覺誘發腦電波(steady-state visual evoked potential, SSVEP)訊號做為輸入訊號，利用經驗模態分解法與refined generalized zero-crossing (rGZC)萃取與辨識短時間腦波訊號，經由EMD與rGZC方法辨識穩態視覺誘發電位，有效率的提升視覺腦波大腦人機介面的正確性與辨識速度，增加訊息傳輸率(information transfer rate, ITR)。在感覺運動區mu波萃取應用中，本研究開發以經驗模態分解法為基礎之單一試驗(single-trial)的腦波律動萃取方法。由於事件相關非同步(Event-related desynchronization, ERD)與同步(synchronization, ERS)腦波律動分析方法已經被應用於大腦感覺運動的功能性研究上，已證明感覺運動皮質區Mu波的ERD主要與運動的籌畫和啟動有關，而ERS則主要與運動的停止或抑制有關。利用經驗模態分解法萃取mu波律動，觀察受試者每一次試驗腦波律動的反應，將有助於腦波律動機轉與帕金森氏的臨床診斷的研究。由於嗅覺事件相關電位（OERP）在臨床嗅覺功能診斷具有直接反應的神經活動和高時間解析度率等優點。然而，研究嗅覺誘發電位必須重複次激提升訊雜比與去除自發性慢波。本研究開發了一種基於經驗模態分解方法將時域信號分解成有限數量總和的固有模態函數（IMF分量）。利用頻率與空間分布的特徵選擇嗅覺相關之IMF分量並重建雜訊抑制之嗅覺誘發電位。Electroencephalography (EEG)/Magnetoencephlography(MEG) has been applied for investigation of neuroscience and diagnosis of many neurological disorders. The noninvasive EEG recordings are overlapping potentials from spontaneous brain rhythm, physiological artifact and external interference. Accordingly, extraction task-related signal features is crucial in the field of EEG/MEG signal processing. The aim of this study is to develop an EMD-based approach to extract single-trial EEG signal. The EMD method decomposed an EEG/MEG epoch into various scale of sub-band component called intrinsic mode function (IMF) and customized the recognition of task-related component to suppress the task-unrelated component. This dissertation evaluates the performance of EMD in EEG or MEG signal into the applications of steady-state visual evoked potential based brain computer interface (SSVEP-based BCI), sensorimotor mu rhythm extraction and olfactory event-related potential extraction. To evaluate the performance of SSVEP feature extraction, this study presents an empirical mode decomposition (EMD) and refined generalized zero crossing (rGZC) approach to achieve frequency recognition in SSVEP-based BCI. The EMD-rGZC improves the information transfer rate (ITR).Event-related desynchronization (ERD) and synchronization (ERS) analysis methods have been widely used in studying movement-related sensorimotor functions in human brain. Movement-related ERD is functionally related to motor planning and initialization, while movement-related ERS is related to motor inhibition and motor cortex resetting. This study developed a single-trial brain rhythm analysis method based on EMD method to discover the mechanisms of ERD and ERS in normal subjects and Parkinson’s patients as well, which could be used as a clinical index for diagnosing Parkinson’s patients.In the study of olfactory event-related potential (OERP) feature extraction, we developed an EMD-based approach to extract OERP from multi-channel EEG recordings. The EMD approach decomposes a signal into IMFs by iteratively conducting the sifting process. Dual criteria on frequency and spatial template were adopted to facilitate the selection of OERP-related IMFs and to reconstruct single-trial OERP for inter-trial investigation.
