A Dual Attention-Based Autoencoder Model for Fetal ECG Extraction From Abdominal Signals

Fetal electrocardiogram (FECG) signals contain important information about the conditions of fetus during pregnancy. Currently, pure FECG can only be obtained through an invasive acquisition process, which is life threatening to both mother and fetus. In this study, single-channel ECG from mother’s abdomen are analyzed with aim extracting clean waveform. This a challenging task due very low amplitude FECG, various noises involved in signal acquisition, overlap R waves. To address problem, we propose novel convolutional autoencoder (AE) network architecture learn extract patterns. The proposed model equipped dual attention mechanism, composed squeeze-and-excitation channel-wise (CW) modules, encoder decoder blocks, respectively. It also benefits bidirectional long short-term memory (LSTM) layer. unique combination allows accurately attend abdominal data. Three well-established datasets considered our experiments. results extraction promising confirm effectiveness using modules within deep learning model. suggest that AE where no maternal (MECG) available.