FECG Extraction Using Bayesian Inference and Neural Networks Approximation

Abstract Neural networks have received much attention in the extraction of fetal electrocardiogram signal in recent years. This paper provides a new method to extract the fetal electrocardiogram signal which uses Bayesian inference and neural networks approximation. We suppose there are two signals obtained from thoracic region and abdominal region of a pregnant woman. Due to the characteristics of Bayesian statistics, electrocardiogram signal can be modeled by generalized Gaussian distribution. Neural networks is used to approximate nonlinear function. The distorted thoracic signal is obtained by backpropagation algorithm because neural networks can uniformly approximate any continuous function if there is sufficient number of neurons in the hidden layer. Iterative technique is also used to update the fetal electrocardiogram signal. By using bayesian inference and neural networks approximation, the fetal electrocardiogram signal is achieved. Simulations demonstrate that the effectiveness of our method.