Hybrid deep learning methodology for fetal cardiac disease prediction

Monitoring of the fetal cardiac cycle is of utmost importance during pregnancy for ensuring fetal health status in a non-invasive way. Irregular fetal cardiac activity indicates heart disease, and that can be treated with antiarrhythmic drugs if diagnosed in the early stages of pregnancy. To acknowledge the above condition, we propose an optimized hybrid deep learning structure, CNN-LSTM, for fetal cardiac disease prediction from abdomen ECG recording on 1D time series data. The CNN excels at identifying spatial features, while the LSTM is superior at learning temporal features. Multichannel abdomen ECG recordings are taken from an open-source, noninvasive fetal ECG arrhythmia database. Abdomen signals are preprocessed and filtered using wavelet transform and median filtering. Fetal cardiac prediction is performed using CNN, LSTM, and a hybrid CNN-LSTM model. We achieved 98% accuracy, 97.2% precision, 98% recall, and 98% F1-score for the hybrid CNNLSTM model. The proposed hybrid deep learning 1D CNNLSTM architecture achieves excellent classification accuracy in its state-of-the-art.