[home] [Personal Program] [Help]

---

Session: Telemedicine - and Fetal ECG
Session starts: Friday 25 January, 13:00
Presentation starts: 14:00
Room: Lamoraalzaal

---

Rik Vullings (Eindhoven University of Technology)
Massimo Mischi (Eindhoven University of Technology)

Abstract:
Blind source separation (BSS) techniques are widely used to extract signals of interest from a mixture with other signals. For example, BSS techniques have been used to retrieve fetal electrocardiogram (ECG) signals from non-invasive recordings performed on the abdomen of a pregnant mother, although with moderate success. One of the reasons for this moderate success is the typically low signal to noise ratio (SNR) of the fetal ECG in the abdominal recordings. As a result, BSS techniques often prioritize other sources, like maternal ECG, muscle activity, and noise, over the fetal ECG [1]. This lack of robustness against poor SNR can be (partly) overcome by incorporating a priori knowledge on the mixing of the source signals. Particularly for electrocardiographic signals, knowledge on the mixing process is available in terms of existing models of the origin and propagation properties of the ECG signals. Unfortunately, virtually none of the available BSS techniques allows for the incorporation of such a priori knowledge. In this study, a novel source separation method is developed that combines the potential accuracy of BSS techniques, in particular independent component analysis (ICA), with the robustness of an underlying physiological model of the electrocardiogram (ECG). The method is developed within a probabilistic framework and yields an iterative estimation of the separation matrix towards a maximum a posteriori solution, where in each iteration the separation matrix is corrected towards the physiological model. The degree of correction is hereby governed by the SNR, providing a flexible trade-off between robustness and accuracy; In case of low SNR, the physiological model contributes more to the estimate of the separation matrix. In case of high SNR, the ICA contributes more to the estimate of the separation matrix. The developed source separation method is evaluated by comparing its performance to that of FastICA [2] on both simulated and real multi-channel ECG recordings. The simulated recordings provide a means for quantitative evaluation and show a 20% decrease in the mean squared error between original ECG sources and retrieved sources for the developed method, with respect to FastICA. As for real data, source separation has been applied on abdominal fetal ECG recordings with the goal of finding a fetal ECG source of sufficient SNR to allow for fetal heart rate detection; The developed source separation method succeeded, while the BSS method did not. Future work includes more extensive evaluation of the developed method as well as further enhancement of the retrieved fetal ECG to allow for the use of fetal ECG in obstetrical diagnostics. REFERENCES [1] S. Harmeling, F. Meineck, K.-R. Müller, “Analysing ICA components by injecting noise”, 4th International Symposium on ICA and Blind Signal Separation, pp. 149-154, 2003. [2] A. Hyvärinen, “Fast and robust fixed point algorithms for independent component analysis”, IEEE Trans on Neural Networks, Vol. 10(3), pp. 626-634, 1999.