Dual-Path Neural Networks Enabled Fetal ECG Extractions From Abdominal Electrohysterograpgy

Fetal heart monitoring is crucial for the early detection of potential risks to fetal health conditions. However, the non-invasive fetal heart monitoring technology using maternal abdominal electrocardiography (AECG) is susceptible to interference from maternal electrocardiography (MECG) and multi-source noise. To address the above challenge, we propose a novel Dual-path Encoder-Decoder Model to extract fetal electrocardiography (FECG) signals from maternal abdominal recordings. It can effectively separate and remove MECG interference to obtain FECG. Additionally, a Feature Augmentation Module based on attention-guided is introduced to optimize the quality of the remaining features after MECG removal. Finally, the optimized features are fed into a decoder with external skip connections to reconstruct the FECG signal. Experimental results show that the proposed method achieves outstanding performance on two public datasets, providing an effective solution for fetal health monitoring and assisting physicians in diagnosing fetal abnormalities.