Study On FECG Extraction Based On Independent Component Analysis Method

The fetal electrocardiogram (FECG) signal contains information on the health of the fetus, and therefore, it has important clinical significance to extract the pure fetal ECG. The fetal ECG in ECG signals recorded at the abdominal surface is very weak, and contaminated by various sources of interference such as maternal electrocardiogram (MECG), the maternal electromyogram (EMG), power line interference and baseline wander. Therefore, how to extract pure fetal ECG signal has become a hot issue.Independent Component Analysis method is an analytical method based on the statistical independence of the source signals. It has been widely used in the FECG extraction. The gradient method or Newton's iterative method is often taken as the optimization approach for searching the separation matrix in traditional ICA methods. They make ICA methods easily strap into local optimum values. The extracted FECG is often mixed with maternal ECG components, and the sequence of the separated signals is uncertain. Therefore, the research of the improved algorithm for ICA methods is done.The main work and fruits are as follows:1. Choosing a changed kurtosis function as the objective function of ICA method, and choosing genetic algorithm as the optimization algorithm of ICA method, a new FECG extraction method is proposed to ensure convergence to the global optimal value as much as possible.2. The genetic algorithm may early converge in the vicinity of the optimal solution. In order to improve this defect, the genetic algorithm and the iteration core of fast fixed-point algorithm are combined for optimization. This method strengthens the optimization capacity.3. In order to eliminate the affect of the noises such as the power-line interference and baseline wander in the extracted FECG, the wavelet packets technique is used.4. The simulation and analysis are done for this method. The simulation results show that:1) The convergence performance of this method is better than the method using genetic algorithm alone for optimization. And the separation accuracy is higher than the traditional ICA methods. There is almost no maternal ECG in the extracted FECG. 2) This method can ensure that the signals can be extracted in order according to the values of kurtosis.3) This method can be used when FECG is mixed with super-Gaussian and sub-Gaussian noises. And noises such as baseline wander and power-line interference can be well suppressed after wavelet packet de-noising.