An efficient unsupervised fetal QRS complex detection from abdominal maternal ECG

Abstract

Non-invasive fetal heart rate is of great relevance in clinical practice to monitor fetal health state during pregnancy. To date, however, despite significant advances in the field of electrocardiography, the analysis of abdominal fetal ECG is considered a challenging problem for biomedical and signal processing communities. This is mainly due to the low signal-to-noise ratio of fetal ECG and difficulties in cancellation of maternal QRS complexes, motion and electromyographic artefacts. In this paper we present an efficient unsupervised algorithm for fetal QRS complex detection from abdominal multichannel signal recordings combining ICA and maternal ECG cancelling, which outperforms each single method. The signal is first pre-processed to remove impulsive artefacts, baseline wandering and power line interference. The following steps are then applied: maternal ECG extraction through independent component analysis (ICA); maternal QRS detection; maternal ECG cancelling through weighted singular value decomposition; enhancing of fetal ECG through ICA and fetal QRS detection. We participated in the Physionet/Computing in Cardiology Challenge 2013, obtaining the top official scores of the challenge (among 53 teams of participants) of event 1 and event 2 concerning fetal heart rate and fetal interbeat intervals estimation section. The developed algorithms are released as open-source on the Physionet website.

Non-invasive fetal ECG analysis

“…Varanini et al (2014) removed the MECG from the abdominal signal using a PCA-based template subtraction algorithm and then applied ICA on the residuals. Then they selected one of the residual based on: knowledge of typical FHR, mean of absolute RR first derivative and mean of absolute RR second derivative and the number of detected FQRS.…”

“…The most significant eigenvectors were fitted back to individual wave epochs from the MECG in order to remove them. The approach for suppressing the MECG is similar to Varanini et al (2014) and Behar et al (2014c) although Lipponen and Tarvainen (2014) separated the MECG cycles into P, QRS and T-waves.…”

Extract Fetal ECG from Single-Lead Abdominal ECG by De-Shape Short Time Fourier Transform and Nonlocal Median

“…Most algorithms need multiple aECG channels and/or one maternal thoracic ECG (mtECG) signal, or at least one aECG channel and one mtECG; including: blind source separation (BSS) [6][7][8][9], semi-BSS like periodic component analysis (πCA), or πTucker decomposition, which takes the pseudo-periodic structure into account [10][11][12], echo state neural network [13], least mean square (LMS) [14], recursive least square (RLS) [13], and blind adaptive filtering [15], Kalman filter [16][17][18], channel selection approach based on features extracted by different methods, like discrete wavelet transform [19], timeadaptive Wiener-filter like filtering [20], principal component regression [21], phase space embedding [22], to name but a few. On the other hand, fewer algorithms depend on the singlelead aECG signal; e.g., template subtraction (TS) [13,[23][24][25][26], and its variation based on singular value decomposition (SVD) or principal component analysis [27,28], the time-frequency analysis, like wavelet transform, pseudo-smooth Wigner-Ville distribution [29][30][31][32] (in practice, three aECG channels are averaged in [30]), and S-transform [33], sequential total variation [34], adaptive neuro-fuzzy inference system and extended Kalman filter [35], particle swarm optimization and extended Kalman smoother [36] state space reconstruction via lag map [37,38], etc.…”

A Combined Independent Source Separation and Quality Index Optimization Method for Fetal ECG Extraction from Abdominal Maternal Leads

“…However, TS is usually implemented by first estimating the mean contribution of the maternal cycle (template) and then subtracting it from the mixture of abdominal signals [18,19]. A powerful method to estimate the contribution of each single maternal heart beat to the abdominal signal is to use a reduced space approximation by Principal Component Analysis (PCA) which can be implemented by Singular Value Decomposition (SVD) [20,21,22,23]. …”