Fetal QRS complex detection using semi-blind source separation framework

Fetal heart rate variability (FHRV) is one of the valuable features of fetal electrocardiography that could be useful to obtain reliable information about the fetal heart activity. In noninvasive systems, the major obstacle for the accurate detection of the fetal QRS (fQRS) complex is the presence of abdominal noise and the maternal ECG (mECG). In this study, we proposed a multistep framework based on semi-blind source separation (SBSS) technique to separate the ECG sources. These sources are ranked automatically to determine the cardiac channel reference and their ECG R-peaks. Using the peaks, the maternal artifacts are reduced from the signals utilizing the quasi-periodicity of cardiac activities. The fetal ECG is also estimated by applying a SBSS stage to the residue of the previous stage. Another channel selection technique, based on a measure of variance, is applied to the channels to rank the channels according to their resemblance with the fetal ECG. The channel most resembling the fQRS is finally used in the peak detection algorithm for finding the fQRS complexes. The procedure was applied to the CinC challenge 2013 dataset B, which consists of 100 noninvasive signals containing the mECG, fECG, and some abdominal noises. The FHR (Entry 4) and RR interval (Entry 5) time series was calculated respectively with obtained average scores 210.12, and 21.23.