| 摘要註: |
人的一生大概有1/3的時間是睡眠,而睡眠大致可分為兩個階段:非快速眼動期(NREM)和快速眼動期(REM),在睡眠過程中,大腦是持續在活動的,經研究發現指出,不同階段的大腦活動與疾病或是學習有關,因此可知睡眠階段判讀的重要性;本論文的目標就是在隱藏式馬可夫模型(Hidden Markov Model, HMM)的主架構下,依據腦電波圖(EEG)、眼電圖(EOG)及肌電圖(EMG)信號,提出一個值得專家與醫師信賴的自動睡眠判讀方法,來提升睡眠醫療診斷的準確性與效率。 在研究中,我們藉由隱藏式馬可夫模型(HMM)結合基因演算法(Genetic Algorithm , GA)和模糊向量量化(Fuzzy Vector Quantization, FVQ)的方法,來提升睡眠判讀的辨識率,其中,基因演算法是用來訓練出較佳的碼簿,而模糊向量量化是用來調整HMM模型;此外,在腦電波睡眠特徵的選取方面,除了根據1968年所提出的R&K規則來選取最具代表性的特徵之外,也選用其他具有輔助性的特徵,全部作為隱藏式馬可夫模型(HMM)訓練的資料來源;與其他使用HMM作為睡眠判讀主要方法的論文不同處,在我們的實驗中,會根據睡眠狀態轉移的特性去調整HMM的模型。由實驗結果可知本論文所提出的方法與現有的研究比較起來,辨識率大大的提升。 Over the span of a lifetime, humans spend about 1/3 of it on sleep. Sleep can be divided into two stages: non rapid eye movement (NREM) and rapid eye movement (REM). During sleep, the brain is continuously active. Studies have found some correlations between brain activities during different stages of sleep and diseases or learning. From these basis, we realize the importance of sleep stage recognition: to bring forth a method of automatic sleep staging based on EEG, EOG and EMG, one that can be relied on and trusted by doctors and experts, using the main frameworks of the Hidden Markov Model (HMM). With this method, the accuracy and efficiency of sleep medical diagnosis can be expected to improve. In our research, we used HMM combined with Genetic Algorithm (GA) and Fuzzy Vector Quantization (FVQ) to improve the recognition rate of sleep staging. The GA is used to train a better codebook for HMM while the FVQ is used to model HMM to improve the performance of the HMM. In addition to selecting characteristic sleep features of EEG based on 1968's R&K rules, features used in other research are also collected as supporting features in this thesis. All features are fed into the HMM's training model. Unlike the existing research on sleep staging by HMM, in which the modeling of HMM is independent of the special properties of the sleep stage transition, the HMM in this study is adjusted to meet the special properties. The results of the experiments show that the proposed method greatly enhances the recognition rate compared with other existing research. |