Extraction of the fetal ECG in noninvasive recordings by signal decompositions

摘要

No signal processing technique has been able to reliably deliver an undistorted fetal electrocardiographic (fECG) signal from electrodes placed on the maternal abdomen because of the low signal-to-noise ratio of the fECG recorded from the maternal body surface. As a result, this led to increased rates of Caesarean deliveries of healthy infants. In an attempt to solve the problem, Physionet/ Computing in Cardiology announced the 2013 Challenge: noninvasive fetal ECG. We are suggesting a method for cancellation of the maternal ECG consisting of: maternal QRS detection, heart rate dependant P-QRS-T interval selection, location of the fiducial points inside this interval for best matching by cross correlation, superimposition of the intervals, calculation of the mean signal of the P-QRS-T interval, and sequential subtraction of the mean signal from the whole fECG recording. Three signal decomposition methods were further applied in order to enhance the fetal QRSs (fQRS): principal component analysis, root-mean-square and Hotelling's T-squared. A combined lead of all decompositions was synthesized and fQRS detection was performed on it. The current research differs from the Challenge in that it uses three signal decomposition methods to enhance the fECG. The new results for 97 recordings of test set B are: 305.657 for Event 4: Fetal heart rate (FHR) and 23.062 for Event 5: Fetal RR interval (FRR).