Foetal electrocardiography (FECG) plays an important role in prenatal monitoring. However, the abdominal electrocardiography (AECG) recorded at the maternal abdomen is significantly affected by the maternal electrocardiography (MECG), making the extraction of FECG a challenging task. This paper presents a deep learning method for MECG removal from single-channel AECG. Firstly, the short time Fourier transform (STFT) is applied to obtain the two-dimensional (2D) features of AECG in time-frequency domain. Secondly, the Convolutional Auto-encoder (CAE) is used to estimate the 2D features of MECG. Finally, after subtracting the estimated MECG, the FECG can be extracted from the AECG. Unlike the methods eliminated the MECG in the 1D time domain, the proposed method focuses on estimating the MECG in the 2D time-frequency domain, where we can take advantage of the structured information in the ECG data. Experimental results on two FECG databases show that the proposed method is effective in eliminating the features of MECG. This study facilitates the clinical applications of FECG in the foetal monitoring.
Paper
Full text
Maternal ECG removal using short time Fourier transform and convolutional auto-encoder
Semantic Scholar · Medicine · 2020
Abstract
Foetal electrocardiography (FECG) plays an important role in prenatal monitoring. However, the abdominal electrocardiography (AECG) recorded at the maternal abdomen is significantly affected by the maternal electrocardiography (MECG), making the extraction of FECG a challenging task. This paper presents a deep learning method for MECG removal from single-channel AECG. Firstly, the short time Fourier transform (STFT) is applied to obtain the two-dimensional (2D) features of AECG in time-frequency domain. Secondly, the Convolutional Auto-encoder (CAE) is used to estimate the 2D features of MECG. Finally, after subtracting the estimated MECG, the FECG can be extracted from the AECG. Unlike the methods eliminated the MECG in the 1D time domain, the proposed method focuses on estimating the MECG in the 2D time-frequency domain, where we can take advantage of the structured information in the ECG data. Experimental results on two FECG databases show that the proposed method is effective in eliminating the features of MECG. This study facilitates the clinical applications of FECG in the foetal monitoring.