SS-SWT and SI-CNN: An Atrial Fibrillation Detection Framework for Time-Frequency ECG Signal

Abstract

Atrial fibrillation is the most common arrhythmia and is associated with high morbidity and mortality from stroke, heart failure, myocardial infarction, and cerebral thrombosis. Effective and rapid detection of atrial fibrillation is critical to reducing morbidity and mortality in patients. Screening atrial fibrillation quickly and efficiently remains a challenging task. In this paper, we propose SS-SWT and SI-CNN: an atrial fibrillation detection framework for the time-frequency ECG signal. First, specific-scale stationary wavelet transform (SS-SWT) is used to decompose a 5-s ECG signal into 8 scales. We select specific scales of coefficients as valid time-frequency features and abandon the other coefficients. The selected coefficients are fed to the scale-independent convolutional neural network (SI-CNN) as a two-dimensional (2D) matrix. In SI-CNN, a convolution kernel specifically for the time-frequency characteristics of ECG signals is designed. During the convolution process, the independence between each scale of coefficient is preserved, and the time domain and the frequency domain characteristics of the ECG signal are effectively extracted, and finally the atrial fibrillation signal is quickly and accurately identified. In this study, experiments are performed using the MIT-BIH AFDB data in 5-s data segments. We achieve 99.03% sensitivity, 99.35% specificity, and 99.23% overall accuracy. The SS-SWT and SI-CNN we propose simplify the feature extraction step, effectively extracts the features of ECG, and reduces the feature redundancy that may be caused by wavelet transform. The results shows that the method can effectively detect atrial fibrillation signals and has potential in clinical application.

Detection of Brief Episodes of Atrial Fibrillation Based on Electrocardiomatrix and Convolutional Neural Network

“…In this sense, increased computational power and the availability of ECG databases with clinical annotations have driven the development of DL techniques for unsupervised ECG analysis (Parvaneh et al, 2019 ; Somani et al, 2021 ). For the detection of AF different DL methodologies have been proposed, including hierarchical attention networks (Mousavi et al, 2020 ), long short-term memory (Faust et al, 2018 ; Andersen et al, 2019 ; Dang et al, 2019 ; Jin et al, 2020 ), convolutional neural network (CNN) (He et al, 2018 ; Xia et al, 2018 ; Lai et al, 2019 ; Huang and Wu, 2020 ; Zhang et al, 2020 ), and approaches combining recurrent neural networks with CNN (Fujita and Cimr, 2019 ; Shi et al, 2020 ; Wang, 2020 ). Some of these approaches are trained with raw ECG signals (Dang et al, 2019 ; Huang and Wu, 2020 ; Jin et al, 2020 ; Mousavi et al, 2020 ; Shi et al, 2020 ; Wang, 2020 ), or with series of RR intervals (Faust et al, 2018 ; Andersen et al, 2019 ; Dang et al, 2019 ; Lai et al, 2019 ), while some others utilize time-frequency domain information extracted from the ECG.…”

“…Some of these approaches are trained with raw ECG signals (Dang et al, 2019 ; Huang and Wu, 2020 ; Jin et al, 2020 ; Mousavi et al, 2020 ; Shi et al, 2020 ; Wang, 2020 ), or with series of RR intervals (Faust et al, 2018 ; Andersen et al, 2019 ; Dang et al, 2019 ; Lai et al, 2019 ), while some others utilize time-frequency domain information extracted from the ECG. For the latter, different transformations have been used to create time-frequency images from the ECG such as the spectrogram (Xia et al, 2018 ), the scalogram (He et al, 2018 ; Jin et al, 2020 ), and the stationary wavelet transform (Xia et al, 2018 ; Zhang et al, 2020 ).…”

“…One drawback of some DL approaches for AF detection is that they are trained and tested with intra-patient subsets of the same database, increasing the chances of overfitting (He et al, 2018 ; Xia et al, 2018 ; Dang et al, 2019 ; Fujita and Cimr, 2019 ; Lai et al, 2019 ; Huang and Wu, 2020 ; Mousavi et al, 2020 ; Shi et al, 2020 ; Wang, 2020 ; Zhang et al, 2020 ). In this sense, further approaches should consider to train, validate, and test the hyperparameters in the network following a inter-patient approach, i.e., perform the validation and testing with records that have not been used during the training process.…”