Independent Component Analysis With Data-centric Contrast Functions For Separating Maternal And Twin Fetal ECG

Intra-uterine fetal ECG monitoring is critical for identical twin gestations because of the increased risk for cardiac defects. Ultrasound echocardiography is important clinically but it does not provide entirely conclusive information pertaining to the fetal cardiac conduction system. On the other hand, fetal electrocardiography can be obtained indirectly by means of ordinary electrodes placed on the mother’s abdomen, allowing potential separation of the fetal electrocardiogram (FECG) for determination of fetal ECG heart rate and morphology. Difficulties with the twin FECG separation is due to the large maternal ECG interference, surrounding noise, artifacts and the underlying similar twin ECG morphology, amplitudes and heart rates. The objective of this work is to investigate Fast-ICA, a signal separation technique, using standard contrast functions and a newly optimized data centric method in this context under different types of interference. We clearly show with a variety of simulations that the chosen polynomial based contrast functions (3 − 6 order) perform superior to the data based Pearson method. They work on par and in some cases better than standard ICA polynomial schemes like SKEW and POW3. Similar trends were seen with a sample of in-vivo data as well. Data-centric schemes used on the ICA have significant potential in true in-vivo situations where separation is based on the underlying data characteristics.