Semi-Blind Source Separation for fetal ECG extraction

The thesis deals with the extraction of the fetal Electrocardiogram (fECG) signal from non-invasive ECG measurements recorded from the abdomen of a pregnant woman. As in adults, the well-being of a fetus can be assessed from the fetal ECG (fECG) signal.
Abdominal ECG recordings can be considered as a summation of different contributes: the maternal ECG (mECG) signal, the fECG signal and interference signals like power line interference, electromyography signals, motion artifacts, etc. It would be extremely useful to find a method for reliably extract the fECG signal. Many authors have been interested to this problem and nowadays various promising methods, mainly based on signal processing, were presented. However, due to the complexity of the problem none of them provided definitive results.
Extraction of fECG from a signals mixture suggest the application of independent component analysis (ICA), also because the assumption that the sources in the measurements (fECG, mECG, other noises) are independent is well verified at the short time scale used for separation. Indeed, Independent Component Analysis or Blind Source Separation (BSS) method were developed to separate mixtures of sources without any a-priori information. However, known limitation of this technique concern the number of recording needed to effectively separate signals and statistical property of the signals. ICA has several variants, in particular fastICA and JADE algorithms were considered in previous studies about this research topics.
In this thesis I proposed a methodology to overcome the limitation of ICA signal processing mainly the need of a high number of abdominal recordings, which limit the clinical application of the methodologies.
I tried to use all the a priori information about the interferences superimposed to the fECG (i.e. maternal ECG, baseline wander, power line interference, etc.) in order to preprocess abdominal recording before applying ICA separation algorithms. Therefore, the proposed method took the name of Semi-Blind Source Separation.
Real abdominal recording from pregnant women were employed to test effectiveness of the proposed methodology.
Experimental results shown that it is possible to obtain reliable fECG information (both heart rhythms and signal morphology) using only four abdominal leads.