Multi stage Principal Component Analysis based method for detection of fetal heart beats in abdominal ECGs

Electrical activity of fetal heart is present in registered abdominal signals, however in real clinical recordings energy of this signal is so small that classical multivariate analysis based methods (e.g. Independent Component Analysis) fail to extract it for further analysis. Principal Component Analysis based method was used for truncated representation and subtraction of cardiocycles of maternal ECG. The energy of fetal ECG component then remains sufficient when compared to other components of the remaining signal to be concentrated in the first component obtained during the second stage Principal Component Analysis. This stage was performed not for extracted cardiocycles, but for the whole leads of the signal. The procedure extracts maximal amplitude of the sought signal regardless to the orientation of the fetus in multilead recording cases. Two stage detection of fetal heart beats is performed in band-pass filtered first principal component. Maximization of correlation with sliding QRS model gives the final time point of fetal heart beat in the region preliminary detected by means of simple amplitude thresholding. Third stage Principal Component Analysis was used to reconstruct the cardiocycles of fetal ECG for further analysis. The algorithm was tested with PhysioNet Challenge 2013 signals. Preliminary results of the detection of fetal heartbeats by our method for the PhysioNet Challenge 2013 on set B were: Average Score (event 1/4): 1391.70; Average Score (event 2/5): 45.49.