Noninvasive fetal QRS detection using Echo State Network

The proposed method combines established cardiology-specific techniques based more on domain knowledge with powerful supervised general-purpose machine learning approaches that are more data-driven. After filtering and normalization, maternal QRS complexes are detected and averaged maternal ECG is removed. The key task of detecting fetal QRS complexes is performed by an Echo State recurrent neural Network (ESN) trained by supervised machine learning. The training of the model is made possible by the availability of correctly annotated training data. Finally, fetal QRS annotations are obtained by a statistics-based dynamic programming approach interpreting the outputs of the ESN. The proposed approach is quite generic and can be extended to other type of signals and annotations.