Hybrid Method Based on Extreme Learning Machine for Fetal Electrocardiogram Extraction

Abstract

A novel hybrid method on Fetal Electrocardiogram (FECG) extraction, which combined FastICA, Extreme Learning Machine (ELM) and adaptive comb filter (ACF), was proposed. Firstly, the baseline drift and other noise in raw maternal abdominal signals were suppressed utilizing conventional filtering method. Then the maternal electrocardiogram (MECG) estimation and FECG estimation containing residual MECG component were obtained from multi-channel maternal abdominal signals by FastICA. The non-linearly transform between MECG estimation and residual MECG component was fitted using ELM. By MECG estimation undergoing the non-linearly transform, the optimal estimation of residual MECG component was obtained. The noisy FECG was extracted by suppressing the estimated MECG component. At last, the FECG enhancement was performed utilizing ACF. The proposed FECG extraction method was evaluated on clinical data. The $\text{SNR}_{\text{svd}},\text{SNR}_{\text{cor}}$, Se, PPV and $\mathrm{F}_{1}$ score of proposed hybrid FECG extraction method are 7.6560dB, 7.8415dB, 99.20%, 98.41% and 98.80%, respectively. The experiment results indicated it better than other conventional method.