Fetal ECG Extraction Using Joint Sparse Supports and a Multiple Measurement Vector (MMV) Approach

Fetal ECG Extraction Using Joint Sparse Supports and a Multiple Measurement Vector (MMV) Approach

Internship Description

​Electrocardiogram (ECG) signals are vital tools in assessing the cardiac health of the mother and of the developing fetus during the course of pregnancy [1]. While the mother's ECG (MECG) recordings is a straight forward procedure using electrodes attached to the mother's chest, obtaining the fetal ECG (FECG) recordings is a much more complex procedure. This is mainly because while the ECG recordings obtained from the mother's chest (called thoracic ECG signals) represent the ECG signal purely from the mother herself, the fetus ECG signals need to be extracted from the ECG recordings obtained from the electrodes placed at the abdomen.  Recordings from these abdominal electrodes comprises of two main components, the MECG and FECG, both superimposed on each other. The FECG component is much smaller in magnitude compared to the MECG part. Moreover, the recorded signals are also affected by distortions and noise from the sources. Separating the mother's ECG signal from these abdominal recording and extracting the fetus ECG component in the presence of noise is an important problem that has been widely studied. Numerous approaches have been proposed in the literature such as adaptive filtering[2], blind sources separation [5, 6], independent component analysis[7, 8], approaches based on sparse redundant representations [4, 3] and Wavwlets [9, 10, 11]. In [2], the authors proposed method to cancel out the effects of the MECG signal component in the abdominal recording using a reference signal from the mother's chest and an adaptive filtering mechanism whose weights are determined using the recursive least squares algorithm. In [5], a blind source seperation mechanism was investigated exploiting the cyclo-stationarities in the ECG signals as useful prior information in the FECG extraction process. The use of sparse redundant representation using learned dictionaries was investigated in [4]. The suggested approach was based on spatially filtering abdominal ECG recordings treating fetal ECG component as noise to obtain the pure MECG component, and then finally subtracting this component  from the original adbominal recordings to obtain the FECG component. The authors in [3]  proposed the use of multi-component dictionaries taking into consideration different structural parts of the ECG signals. An adaptive filtering approach based on wavelet representation of the ECG signal was proposed in [9]. In this work, we will pursue and new approch for extracting the FECG signals from the abdominal ECG recordings. The proposed approach will use sparse domain reresentation based on multi-component wavelet dictionaries combined with multiple measurement vectors techniques.


Faculty Name

Field of Study

​Electrical Engineering