請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/42846
標題: | 使用獨立成分分析在多通道壓抑之母親心電圖實現胎兒心電圖粹取 Fetal ECG Extraction using ICA of Multi-Channel Maternal ECG Suppression |
作者: | Chao-Yang Hsu 許朝陽 |
指導教授: | 曹建和 |
關鍵字: | 胎兒心電圖,信號處理,PQRST波樣版,獨立信號分析,投影量追蹤, fetal electrocardiogram,signal processing,template PQRST complex,independent component analysis,projection pursuit, |
出版年 : | 2009 |
學位: | 碩士 |
摘要: | 心電圖,是心臟檢查中一種十分重要,並且普及的非侵入式方法。藉由分析在皮膚上的電極所紀錄的電訊號,可以觀察出心臟機能是否異常。同理,胎兒的心臟所產生之電訊號,也可以經由孕婦皮膚上的電極所紀錄,觀察胎兒的心臟機能,稱之為胎兒心電圖。如同標準心電圖,胎兒心電圖也同時代表著其他的生理資訊,如經由連續心跳速率的變化來觀察胎兒的自主神經系統。不同的是,胎兒心電圖的觀察卻有實際上的困難。因為胎兒位於母體之內,皮膚上的電極所紀錄的信號中,同時存在兩個本質上相同的來源,為母親和胎兒的心臟。尤其母親心電圖的信號強度遠大於胎兒心電圖,更增加了處理上的困難。另外,因為胎兒心電圖十分微弱,其他生理現象所產生的干擾或是量測上造成的雜訊,相對於胎兒心電圖的影響也會十分顯著。本論文研究著力於胎兒心電圖的信號取得。本論文的目的為提出一個新架構之演算法,在不需要胸部電極的條件下,只經由少數腹部電極的信號,能夠正確地粹取出胎兒心電訊號,提供更進一步的胎兒生理資訊分析。本論文所提出的演算法整合了不同面向的處理,包括頻率領域,時間領域,空間領域和統計領域。首先的帶通濾波器,利用信號間頻率成分的差異,初步去除其他的干擾和雜訊,保留母親和胎兒的心電訊號成分。接下來將母親的心電圖視為一連續的PQRST波的組成,可以藉由PQRST波的樣板,在每一個時間基準點上重建一估計的母親心電訊號。而母親心電圖的去除,則將經過帶通濾波器處理之腹部電極的信號,減去此估計的母親心電訊號來實現。由於真實心跳的時變性,甚至相鄰的PQRST波也很難完全相同,造成使用直接減去樣版的方法,將無法將母親心電圖完全的去除。所以接著使用屬於獨立成分分析中的投影量追蹤方法,進一步的去除剩餘的母親心電圖和其他雜訊,利用信號的統計特性來粹取出胎兒心電圖。實驗的信號使用公開在PhysioBank資料庫上的『Non-Invasive Fetal Electrocardiogram Database』,由西班牙之瓦倫西亞大學〈University of Valentia〉的Marcelino Martinez Sober博士和Jorge Granado Marco所量測之心電圖信號。實驗的結果顯示本論文所提出之演算法實現了在只有四個腹部電極的條件下,完成胎兒心電訊號的粹取。同時並討論在演算法中的設計參數,對於胎兒的心電訊號粹取所造成的影響,分別為帶通濾波器的截止頻率,母親心電訊號樣版的長度和基準點位置,投影量追蹤方法的處理視窗。經由實驗結果所顯示的胎兒心電訊號,本論文的結論為,將時間領域的母親心電訊號樣版減去法,適當地結合屬於獨立成分分析中的投影量追蹤方法,兩種方法彼此間有互補的效果,所提出的演算法實現了一有效並正確的胎兒心電圖粹取。 Electrocardiography (ECG or EKG) is the important and general non-invasive method for heart diagnosis. The recording of the electrical activity from cutaneous electrodes provides observation of abnormal heart function. Likewise, the recording of the electrical activity from cutaneous electrodes of the pregnant woman can use for observation of fetal heart function, called fetal ECG (FECG). As same as standard ECG, the fetal ECG also represents other biomedical information, such as the observation of fetal autonomic nervous system (ANS) by successive changes of fetal hart rate variability (HRV). Different from this, the observation of fetal ECG has difficulty in practice. Because fetus is stored inside maternal body, there are two natural identical signal sources exist simultaneously, such as maternal and fetal heart in recording signal from skin electrodes. Especially the signal strength of maternal ECG is much large then fetal ECG is increase the difficulty of signal processing. Furthermore, compare with weak fetal ECG, it is very evident that the interference generated by other physiologic phenomenon or measuring noise influence the signal of fetal ECG. The research of this thesis focuses on the signal obtainment of fetal ECG. The objective of this thesis is to propose new algorithm architecture for proper fetal ECG extraction without thoracic electrode, based on few abdominal electrodes only. The processing result can be used for further analysis of fetal medical information. The proposed algorithm integrates several processing by various perspectives, including frequency domain, time domain, spatial domain and statistic domain. The first band-pass filter preliminary removes other interference and noise based on signal composite differences of frequency component. Next, the maternal ECG is represented as a summation of sequential PQRST complexes. The maternal ECG reconstructed by positioning template of PQRST complex on each fiducial point in time. Based on this approach, the output of band-pass filter is subtracted to the estimated maternal ECG for maternal ECG elimination. Due to time-dependent of real heart beat, even neighboring PQRST complex is hardly identical. Therefore the direct subtraction method does not remove maternal ECG entirely. Afterward, the further elimination of residual maternal ECG and other noises achieves by using a kind of independent component analysis (ICA) method, call projection pursuit (PP). It extracts fetal ECG based on statistical characterization of source signal. The experimental signals download from “Non-Invasive Fetal Electrocardiogram Database” of “PhysioBank”, a publish database in internet. Those signals had measured by Ph.D. Marocelino Martinez Sober and Jorge Granado Marco in the University of Valencia, Spain. The experimental results demonstrate proposed algorithm achieves fetal ECG extraction from four recording of abdominal electrodes only. Meanwhile, the discussion of how several design parameters, such as the cut-off frequency of band-pass filter, the length and position of template of maternal PQRST complex, the processing window of projection pursuit impact the fetal ECG extraction is described in this thesis. In conclusion, corresponding experimental results prove the template based maternal ECG subtraction method and ICA based projection pursuit method can be integrated together by following a proper sequence and adjustment, the proposed algorithm succeeds fetal ECG extraction effectively and accurately. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/42846 |
全文授權: | 有償授權 |
顯示於系所單位: | 電機工程學系 |
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 | 2.47 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。