ECG classification using 1-D convolutional deep residual neural network.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cardiology

Background:

• Electrocardiograph (ECG) signals are crucial for diagnosing cardiovascular diseases (CVDs).
• Traditional ECG classification methods are complex and costly due to extensive signal processing.
• Deep learning offers a promising alternative for efficient ECG analysis.

Purpose of the Study:

• To develop a deep learning (DL) system for accurate ECG signal classification.
• To implement a 1-D convolutional deep residual neural network (ResNet) for automated feature extraction from heartbeats.
• To address class imbalance in ECG datasets using the Synthetic Minority Oversampling Technique (SMOTE).

Main Methods:

• Utilized a 1-D convolutional ResNet model for direct feature extraction from ECG heartbeats.
• Employed the SMOTE technique to handle class imbalance in the training data.
• Evaluated the classifier's performance using ten-fold cross-validation (CV) with metrics including accuracy, precision, sensitivity, F1-score, and Kappa.

Main Results:

• Achieved an average accuracy of 98.63% in classifying five types of heartbeats.
• Reported high performance metrics: precision (92.86%), sensitivity (92.41%), specificity (99.06%), F1-score (92.63%), and Kappa (95.5%).
• Demonstrated the effectiveness of the proposed ResNet model, especially with deeper layers, compared to other 1-D CNNs.

Conclusions:

• The proposed DL-based ResNet system provides an efficient and accurate method for ECG signal classification.
• This approach simplifies the traditional complex signal processing, potentially reducing design costs.
• The study highlights the potential of deep learning, specifically ResNet, for advancing cardiovascular disease diagnosis.