Deep Learning-Based Non-invasive Fetal Cardiac Arrhythmia Detection

Non-invasive fetal electrocardiography (NI-FECG) has the possibility to offer some added clinical information to assist in detecting fetal distress, and thus it offers novel diagnostic possibilities for prenatal treatment to arrhythmic fetus. The core aim of this work is to explore whether reliable classification of arrhythmic (ARR) fetus and normal rhythm (NR) fetus can be achieved from multi-channel NI-FECG signals without canceling maternal ECG (MECG) signals. A state-of-the-art deep learning method has been proposed for this task. The open-access NI-FECG dataset that has been taken from the PhysioNet.org for the present work. Each recording in the NI-FECG dataset used for the study has one maternal ECG signal and 4–5 abdominal channels. The raw NI-FECG signals are preprocessed to remove any disruptive noise from the NI-FECG recordings without considerably altering either the fetal or maternal ECG components. Secondly, in the proposed method, the time–frequency images, such as spectrogram, are computed to train the model instead of raw NI-FECG signals, which are standardized before they are fed to a CNN classifier to perform fetal arrhythmia classification. Various performance evaluation metrics including precision, recall, F-measure, accuracy, and ROC curve have been used to assess the model performance. The proposed CNN-based deep learning model achieves a high precision (96.17%), recall (96.21%), F1-score (96.18%), and accuracy (96.31%). In addition, the influence of varying batch size on model performance was also evaluated, whose results show that batch size of 32 outperforms the batch size of 64 and 128 on this particular task.