The Classification of Fetal ECG Arrhythmia by Using Fusion Method of CNN-Transformer

Early detection of fetal cardiac arrhythmia remains a challenging problem due to the complex and non-stationary nature of fetal electrocardiogram (fECG) signals. Traditional approaches often fail to effectively capture both morphological and temporal dependencies present in these signals. In this study, we propose a hybrid deep learning framework that integrates Convolutional Neural Networks (CNN) and Transformer architecture for accurate classification of fetal arrhythmia using non-invasive fECG signals. The proposed model leverages CNN to extract spatial (morphological) features from ECG signal representations, while the Transformer utilizes a self-attention mechanism to capture long-range temporal dependencies in the signal. Short-Time Fourier Transform (STFT) is employed to convert ECG signals into time–frequency representations, enabling effective feature learning. Experiments are conducted on the publicly available Non-Invasive Fetal ECG Arrhythmia Database (NIFEADB), consisting of recordings from 26 subjects. To avoid data leakage, subject-wise data splitting is considered. The proposed hybrid model achieves a classification accuracy of 96.53 %, outperforming standalone CNN and Transformer models. The results demonstrate that combining spatial and temporal feature extraction significantly improves classification performance and provides a promising approach for early diagnosis of fetal cardiac abnormalities.