Non-invasive fetal multilead RR interval determination from maternal abdominal recordings: The Physionet/CinC Challenge 2013

The PhysioNet Challenge 2013 focused to develop accurate algorithms for locating QRS complexes and estimating the QT interval in noninvasive fetal ECG (FECG) signals obtained from noninvasive abdominal recordings in pregnant women. We upgraded our custom automated software used previously. To cope with signal having different frequency contents we derived and analyzed two sets of signals using two different bandpass filters. The mother ECG (MECG) cancellation was performed using p. QRS and T wave templates and with appropriate scaling and shifting of P and T waves due to QT and PQ variability. From the cancelled signal, true fetal beats were extracted with a FQRS filter assuming that the error variability due to deviations of detected beat intervals from the median RR interval should be minimal. By selecting the lead with the smallest error and interpolating missing beats we obtained the final fetal QRS positions. The accuracy of determination of FQRS positions of our algorithm was tested by comparing our derived FQRS sequences with the reference annotations for the learning test set of 75 recordings. For the PhysioNet 2013 Challenge on set B from we obtained an average score 181,2 for fetal HR estimation accuracy, score 10.92 for fetal inter-beat accuracy.