Deep Learning for Fetal QRS-complex Classification in Noninvasive Fetal Electrocardiogram

Fetal QRS-complex generated from an abdominal electrocardiogram (ECG) is an effective noninvasive tool for accurately measuring the fetus well-being during pregnancy. Classification of fetal QRS is significant to avoid fetal heart abnormalities, such as asphyxia, bradycardia, and hypoxia. However, accurately classifying the fetal QRS-complex is hard because the characteristic fetal ECG (fECG) has small amount of signal energy and it has undesired signal components, including recording devices, movement artifacts, and fetal brain activity. The sophisticated fECG signal processing for fetal QRS-complex classification has been used in the literature. However, they mostly use human intervention for feature extraction, leading to time-consume and raising the measurement error. Hence, this study investigates deep learning (DL) that performs automated feature extraction tasks excellently. DL has excellence amount of data scales, which is its ability to automatically learn and improve from data without the need for explicit programming. A sequence-to-sequence learning task was generated using a hybrid four convolution layers as a part of convolutional neural networks and bidirectional long short-term memory (ConvBiLSTM). As the results, the performance is outstanding; the accuracy, sensitivity, specificity, precision, and F1-score were 100%. The observed results show the proposed ConvBiLSTM model is worth evaluating their effectiveness for fetal QRS-complex classification.