AAFA-Net: Adaptive Amplitude-Frequency Attention Network for Fetal Monitoring From Noninvasive Abdominal Recordings
Yong Liao, Xu Wang, Yang Han, Yamei Deng
Abstract
The noninvasive fetal electrocardiogram (FECG) is helpful for fetal well-being monitoring. However, it is difficult to obtain high-quality FECG signals because of the maternal electrocardiogram (MECG) and noise in the abdominal ECG (AECG). To address this problem, an Adaptive Amplitude-Frequency Attention Network (AAFA-Net) is proposed for extracting FECG signals from AECG signals, where the Frequency Encoder-Decoder (FED) module is developed to distinguish the FECG frequency components from AECG signals, and the Amplitude Encoder-Decoder (AED) module is devised to extract FECG amplitude components from AECG signals, while the Window Encoder-Decoder (WED) module is designed to determine the temporal window around the FECG signal. Experiments conducted on the benchmarks show that the proposed AAFA-Net performs better than the state-of-the-art approaches.