A deep Bi-CapsNet for analysing ECG signals to classify cardiac arrhythmia

Area of Science:

• Cardiology and Artificial Intelligence
• Medical Signal Processing
• Deep Learning for Healthcare

Background:

• Electrocardiogram (ECG) is crucial for screening cardiac arrhythmias.
• Accurate feature extraction and classification are essential for automated heart disease prediction.
• Existing deep learning models require improvement for precise arrhythmia classification.

Purpose of the Study:

• To develop and validate a deep bi-directional capsule network (Bi-CapsNet) for accurate cardiac arrhythmia classification.
• To enhance the precision of automated heart disease prediction using ECG data.
• To compare the performance of Bi-CapsNet against traditional deep learning models.

Main Methods:

• ECG signal acquisition and preprocessing (DC drift removal, normalization, filtering, artifact removal).
• Feature extraction using a Deep Ensemble CNN-RNN approach.
• Classification of cardiac arrhythmias using the Deep Bi-CapsNet model.
• Validation on the MIT-BIH arrhythmia database, identifying Normal, RBBB, PVC, APB, and LBBB types.

Main Results:

• The proposed Bi-CapsNet model achieved an overall accuracy of 97.19%.
• Bi-CapsNet outperformed traditional models: CNN (89.87%), FTBO (85%), and Capsule Network (97.0%).
• Comprehensive performance analysis included metrics like precision, F1-score, sensitivity, and specificity.

Conclusions:

• The developed Bi-CapsNet model demonstrates superior performance in classifying cardiac arrhythmias from ECG signals.
• This intelligent deep learning approach offers a significant advancement in automated heart disease prediction.
• The hybrid model provides a highly accurate and effective method for clinical arrhythmia screening.