Stacking Deep Neural Networks to Detect Multiple Types of Cardiac Arrhythmias.

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Heart arrhythmias pose significant risks, necessitating accurate and early diagnosis via electrocardiograms (ECG).
• Existing deep learning models like CNNs, LSTMs, and Transformers have shown promise but have individual limitations in capturing ECG signal features.
• CNNs excel at spatial features, LSTMs at temporal, and Transformers at long-range dependencies, yet none are perfect alone.

Purpose of the Study:

• To develop and evaluate a two-level stacked ensemble framework for enhanced ECG arrhythmia classification.
• To overcome the limitations of individual deep learning models by integrating their strengths.
• To improve the accuracy and reliability of automated arrhythmia detection systems.

Main Methods:

• A two-level stacked ensemble framework was designed, incorporating Deep Neural Networks (DNN), Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and Transformer models as base learners.
• A Multilayer Perceptron (MLP) was utilized as a meta-learner at the second level to optimally weigh predictions from the base models.
• The framework was trained and validated on the MIT-BIH and INCART arrhythmia databases for classifying five categories of heart arrhythmias.

Main Results:

• The stacked ensemble model demonstrated superior performance compared to individual base models and state-of-the-art techniques.
• Achieved a high F-score of 99.79% on the MIT-BIH database and 99.62% on the INCART database.
• The framework effectively fused spatial, temporal, and long-range dependency features for robust arrhythmia classification.

Conclusions:

• The proposed stacked ensemble framework offers a significant advancement in ECG-based arrhythmia classification.
• This approach effectively leverages the complementary strengths of diverse deep learning architectures.
• The findings suggest a promising direction for developing more accurate and reliable automated diagnostic tools for cardiovascular diseases.