Fetal electrocardiogram extraction with artifact suppression using independent component analysis and empirical mode decomposition

Abstract

According to maternal mortality statistics of the world health organization (WHO), every day, 830 women die from the cause of pregnancy and childbirth. Additionally, deaths mostly occur in developing countries, especially in Asia and Africa. In rural areas and poor communities, the maternal mortality ratio is 239 per 100,000 live births. On the other hand, in developed countries, the ratio is 12 per 100,000 live births. The difference between those ratios shows that better prenatal care and diagnosis can decrease the mortality rate significantly. Because of the high cost of the advanced prenatal diagnosis technique such as amniocentesis, chronic villus sampling, etc., the poor people in developing countries cannot access advanced prenatal health care. The fetal electrocardiogram (ECG) can be used to solve the expense problems in diagnosis. The ECG acquisition is not only a noninvasive method but also provides a doctor to diagnose several prenatal diseases. In addition, it is a low-cost diagnosis because the only simple electrodes are needed for the measurement. Moreover, fetal ECG can provide a doctor to find new ways to research the electrical activity of the fetus. Since the fetal heart does not perform the same function of the adult heart, there is no ground truth about the normal characteristic of fetal ECG. However, for the adult, many diseases are diagnosed by ECG, and it is possible that fetal ECG can provide the data in the same way. Thus, the fetal ECG has the benefit both in the diagnosis term and the research term. Currently, for fetal diagnosis, there are several methods or criteria to observe the prenatal abnormalities and the fetus’s health, such as prenatal ultrasound, amniocentesis, and maternal blood pressure. However, most of those methods are invasive, and some of them affect the fetus, negatively. Especially, for the method which provides a reliable result such as amniocentesis for the genetic disorder and percutaneous umbilical cord blood sampling (PUBS) for genetic defects, a big needle that penetrates a placenta, is used to collect the amniotic fluid. To avoid the risk effect of the prenatal diagnosis, the fetal ECG diagnosis is presented. However, different from the adult ECG, the fetal ECG cannot be directly obtained by using the electrodes. Since the fetus is in the amnion, and the electrodes are placed on the abdominal wall, the signals have to pass through the amnion, placenta, and the abdominal wall before they reach to the electrodes. Thus, the fetal ECG is contaminated by several artifacts such as noise, muscular signal, maternal ECG, and fetal hiccup. Finally, the effective extraction is needed for artifacts elimination, and this study presents the independent component analysis (ICA)-based extraction for fetal ECG. From the algorithm’s efficiency verification, the filtering process can get rid of artifacts such as noise and baseline. Moreover, the empirical mode decomposition can separate the partial overlapped fetal and maternal R-peaks which occurs in the signal. For the independent component analysis (ICA), the procedure can extract the fetal ECG, almost perfectly. The results from the dataset Non-Invasive Fetal ECG Database show that the algorithm cannot extract the signal from 5 subjects out of 53 subjects. Furthermore, the algorithm efficiency measured using the dataset Abdominal and Direct Fetal Electrocardiogram Database is 93.58% averagely, and the best efficiency is 100%. The details about efficiency validation will be explained in chapter 3.