A Fast Adaptive LPCA Method for Fetal ECG Extraction Based on Multichannel Signals

Monitoring fetal electrocardiogram (FECG) is crucial for diagnosing potential congenital heart defects and fetal distress during pregnancy. Due to strong interferences and background noise, it is challenging to extract FECG from maternal abdominal electrocardiogram (AECG). In this article, we propose a novel method for extracting FECG using multiple-channel AECGs. Our approach incorporates canonical correlation analysis (CCA) into the local principal component analysis (LPCA). First, the FECG is preliminarily extracted from multichannel AECGs via CCA approach, which greatly suppresses the maternal interference and retains the waveform details of FECG. However, the extracted signal may still suffer from significant noise. Second, a fast adaptive LPCA algorithm is proposed for denoising purpose, which avoids the computationally expensive eigenvalue decomposition of high dimensional covariance matrix. The convergence of the adaptive LPCA is analyzed based on the deterministic discrete-time (DDT) theory. The proposed algorithm is evaluated on both synthetic signals and real recordings. The simulation results demonstrate the robustness of our method against Gaussian noise, the signal to noise ratio (SNR) of extracted FECG achieves 19.65 dB. The experimental results on real-world dataset further illustrate the outstanding performance of our method in terms of computational load and fetal QRS complex detection accuracy with
$F_{1}$
score of 96.55% on ADFECGDB dataset.