Auto-encoder and bidirectional long short-term memory based automated arrhythmia classification for ECG signal. (August 2022)

• Record Type: Journal Article
• Authors: Ramkumar, M.; Sarath Kumar, R.; Manjunathan, A.; Mathankumar, M.; Pauliah, Jenopaul
• Is Part Of: Biomedical signal processing and control. Volume 77(2022)
• Publication Date: 2022-08
• Subjects: Arrhythmia -- Normal sinus beat -- Bidirectional LSTM -- Pacing beat -- Sinus bradycardia; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2022.103826
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: The combination of AE-biLSTM for automated arrhythmia classification. The input ECG signals are pre-processed by DTCWT for removing the baseline. AE-biLSTM method contains an encoder that extracts higher level feature. Decoder output reconstruct ECG signals from higher level features using biLSTM. Finally classifies the 6 heartbeats such as N, AFIB, B, P, AFL, SBR. Abstract: In this manuscript, the combination of Auto- Encoder and Bidirectional long short-term memory (AE-biLSTM) for automated arrhythmia classification is proposed to automatically classify the six kinds of Electrocardiogram (ECG) signals with low cost. Initially, the input Electrocardiogram signals are pre-processed by Dual tree complex wavelet transform (DTCWT) for removing the baseline. The pre-processed ECG signals are given to the combined network of AE-biLSTM. The proposed AE-biLSTM method contains an encoder that extracts higher level feature from the Electro cardiogram arrhythmias signals using bidirectional long short- term memory (biLSTM) network, then a decoder output reconstruct Electro cardiogram arrhythmias signals from higher level features using biLSTM network. Finally, the proposed method accurately classifies the 6 heartbeats types, such as normal (N) sinus beat, atrial fibrillation (AFIB), ventricular bigeminy (B), pacing beat (P), atrial flutter (AFL), sinus brady cardia (SBR). The simulating process is activated in MATLAB. Lastly, the AE-biLSTM method utilize 2 extra databases: (i) new N beat (ii) AFIB beat, which is self-determining of the network's training database. The proposed model attains the better performance of 97.15 % accuracy, 98.33% positive predictive value, 99.43% sensitivity, 96.22% specificity compared to the existing methods, such as Automated arrhythmia classification based convolutional neural networks with long short-term memory networks (CNN-LSTM), and automated arrhythmia classification based deep code features with long short-term memory networks (DCF-LSTM) respectively.