可適性母親心電圖抑制粹取胎兒心電圖

Abstract

心電圖紀錄著心臟的電訊號活動，它能提供非常重要的健康資訊給醫護人員。 心電圖的量測是經由擺放在人體上的電極來偵測出心電訊號，這種非侵入式的診斷方法在臨床上普遍被用來診斷心臟的相關疾病。異常的心跳訊號、生理和心理狀況都能藉由心電圖的變化中獲得珍貴的訊息。胎兒的心電圖和成人的心電圖一樣隱含著重要的健康資訊，因為它們在生理及心理的狀態表現上是相同的。透過放置在母親腹部上的電極，可因此偵測到微弱的胎兒心電訊號。藉由觀察自主神經系統所造成的連續心跳速率變化，就能得到胎兒的生理健康狀況心理狀態。然而在實際上，由腹部電極所錄得的訊號不僅只包含了胎兒的心電訊號，還有母親的心電訊號及其他雜訊所組成。此外母親的心跳振幅強度更是遠大於胎兒的心跳振幅強度，這使得在訊號處理上增加了許多難度。在過去，有不少人提出使用一個母親的心電訊號單模版來扣除每一個母親心電訊號進而抑制母親的訊號。然而在實際上已被濾除了大部分雜訊的腹部訊號，若考慮母親心跳與胎兒心跳沒有部分和完全重疊的情況下，每個母親的心跳圖形其實有著些微的差異，若只單純使用一個單模版來抑制每個不同的母親心電訊號勢必會有一些殘留訊號未被消除。 此篇論文的目的為提出一個新的演算方法，針對每個不同的母親心跳波形產生出一個相對應的可適性波形來抑制強大且多變的母親訊號及濾除掉不必要的雜訊並且能穩定有效地處理懷孕早期的資料。 演算法一開始，腹部電極所記錄的訊號會先被前處理一次，此前處理過程為訊號先通過一個 5~100 Hz 的帶通濾波器濾除不必要的雜訊，接著為了增強訊號對雜訊比這幾個腹部訊號會經過一次奇異值分解分離出一個接著要進一步處理的訊號。透過母親胸部電極訊號來準確且有效地定位出每個在前處理後的訊號的母親心跳位置。接著，每個前處理後的母親心跳被取出再一次作奇異值分解分離出較強的基底。這些較強的基底會和每一個母親的心跳訊號作結合，經運算後產生出獨特的波形並透過相消來抑制每個不同的母親心跳訊號。最後，再通過一個5~100 Hz 的帶通濾波器濾除由相消後所產生的雜訊。經過這些步驟，胎兒心電訊號就能成功地被粹取出，此訊號就可以作更進一步地健康狀況分析。 此新的演算方法更進一步地成功消除母親訊號，從實驗的結果來看也證明了使用可適性波形相減後的效果比使用單模版相減的效果還要好。訊號對雜訊比明顯被提升了，且使用此演算法能穩定且長期地處理懷孕週數22到40週之間的資料。雖然在22週訊號對雜訊比不是很高，主要是因為22週的資料長度都只介於2~5分鐘，這使得較難分離出較強的基底且22週的資料都只有3個腹部電極相對於4個電極來說訊號對雜訊比就會因此而降低。即使如此對於一般人來說還是能很容易地辨別每個胎兒心跳，再加上在22週時也能成功獲得胎兒心率變異度的資訊，這使得醫護人員能及早掌握到更多資訊讓母親和胎兒能更早獲得更好的醫療及照護。 此演算法使用一個胸部電極和三~四個腹部電極〈視資料內容而定〉。而此論文所處理的實驗訊號皆是從PhysioBank資料庫中的非侵入式胎兒心電圖資料庫〈Non-Invasive Fetal Electrocardiogram Database〉中下載所得，這些訊號是由西班牙之瓦倫西亞大學〈University of Valentia〉的Marcelino Martinez Sober 博士和Jorge Granado Marco 所錄製之心電訊號。

Representing the electrical activity of the cardiac, electrocardiogram (ECG) is a common non-invasive technique for diagnosing heart diseases in a clinical environment. By way of analyzing ECG the patient’s physiological and psychological information can be obtained and offered to medical personnel. Similarly, the health condition of the fetus can be acquired as well through analyzing fetal electrocardiogram (FECG) because the characterization of the FECG is the same as that of the adult ECG. The FECG is derived from cutaneous electrodes placed on the pregnant women’s abdomen. Fetal health information can be acquired through calculating the fetal heart rate variability (fHRV), which be regulated by the autonomic nerve system (ANS) and indicates the fetus’ physiological and psychological conditions, and observing the morphology of ectopic heart beats indicating physiological functions. Unfortunately, abdominal electrodes located on the mother’s abdomen record not only the FECG but undesired signals including the maternal ECG (MECG), other bioelectric source interferences and outer noise. Because the magnitude of the FECG is far smaller than that of the MECG and the FECG can be influenced by noise, it is arduous to extract the FECG from abdominal signals recorded from abdomen electrodes. Previous studies have reported that using the single template PQRST theory suppresses the maternal component to acquire the FECG. However, it may be seen that there is a discrepancy in each maternal beat compared with other maternal beats as abdominal signals are filtered noise and maternal beats do not be entirely or partially overlapped by fetal beats. It can be certain that after subtraction some residual components remain using the single template algorithm. In the thesis, a new algorithm is advanced to extract the FECG. The algorithm, which can be steadily executed in early pregnancy for monitoring the FECG, aims at the shape of each maternal beat to generate an adaptive waveform to which that beat is corresponding and then by which that beat is subtracted. First of all, interferences and noises are preliminarily removed. Secondly, each R wave of the abdominal MECG is orientated. Next, through orientation points each maternal beat is extracted from the abdominal signal. Fourthly, these beats are then calculated by the singular value decomposition (SVD) to extract most dominant bases. Adaptive waveforms are formed with these bases and each maternal beat itself. Subsequently, the corresponding adaptive waveforms are used to suppress maternal beats in the abdominal signal. The last band-pass filter then eliminates the noise generated during subtraction. Finally, the FECG signal can be acquired. Of the new algorithm compared with the result of the single template algorithm, the result demonstrates that using the new algorithm the FECG can be more visible and that the signal-to-noise (SNR) can be further enhanced. The result also illustrates that the FECG can be extracted for long-time monitoring. The new algorithm can successfully process experimental data between 22 and 40 weeks of gestation. The experimental data used in the thesis were downloaded from the “Non-Invasive Fetal Electrocardiogram Database” of “PhysioBank” on the internet. Although the SNR of the result at 22 weeks of pregnancy is not high enough, it is effortless to identify the fetal QRS complex. The low SNR is because the length of all of 22 weeks data range from two to five minutes and the number of abdomen electrodes is three. Short length of data enables the algorithm to hardly extract most dominant bases, and three electrodes compared with four have the lower SNR during executing the SVD to separate information and noise. Nevertheless, in different weeks of gestation the FECG can be further analyzed to obtain the fHRV information. It enables medical personnel to know at early stages the health condition of the fetus on which the mother and the fetus can rely to receive proper treatment.