Research On Modeling Of Multi-channel Hybrid ECG And Extraction Algorithm Of Fetal ECG

Congenital heart defects is the most common birth defects which formed in theearly stages of pregnancy, and is also the main cause of newborn deaths. Most heartdefects will be reflected in the electrocardiogram (ECG) which contains morephysiological information than conventional ultrasound methods. Therefore, regularmonitoring of the fetal electrocardiogram (FECG) may help to detect cardiacabnormalities early, so obstetric and pediatric can determine early medication orconsider the necessary measures to prevent and resolve the problem after birth orduring childbirth. However, the fetal ECG is very weak, also vulnerable to theexternal environment and the mother’s own interference. That’s the main problem toextract pure fetal ECG.In this paper, the generation principle of fetal ECG signal, ECG noise model,the dipole vector model and multi-channel ECG mixed model are researched.Especially, the paper focuses on the estimation algorithms which can optimize theparameters of the model mentioned above and also researches spatial filteringalgorithm of synthetic analog mixed ECG and extraction algorithm of fetal ECG.The main contents include the following aspects.Firstly, synthesize the ECG analog noise. According to AR model which is theactual ECG noise model, train the model with the real ECG noise databases basedon Kalman filtering algorithm and optimize the model parameter. Synthesis of theactual ECG analog noise signal with different noise-signal ratio provide theoreticalbasis of synthetic of real ECG and optimization of the electrode arrangement onmaternal abdomen.Then, synthesize the mixed multi-channel analog ECG of maternal abdomen.Parameter optimization of dimensional dynamic dipole vector model is based onleast squares gradient descent algorithm. R-wave detection, heart beat slicing andcharacteristic parameters extraction help to achieve the initialization of vectordipole model. The mean square error can evaluate the accuracy of the model. Bodyvolume conductor model formula is deduced according multi-channel ECG model.ECG synthesized with non-invasive ECG database composite with synthetic analognoise will output mixed analog ECG of maternal abdomen. Next, filter noise of ECG by EKF and EKS algorithms. Linearized model ofthe fifth-order Gaussian kernel function based on form of polar coordinates andestablish the state equation and observation equation. Use3σ guidelines to estimatedistribution of error of the two algorithms roughly, and the introduction of standardnoise-signal ratio provide a more precise evaluation of the effectiveness of twofiltering algorithms.Finally, extract the fetal ECG by blind source separation algorithm. Separatethe fetal ECG using multi-channel ECG model based on blind source separationalgorithm. Amendatory output SNR can compare two extraction algorithmscontaining Fast ICA algorithm and JADE algorithm. The results show that Fast ICAalgorithm converges faster, while JADE algorithm suppresses noise stronger and isable to extract the fetal ECG clearer.