ECG-Based Multiclass Arrhythmia Classification Using Beat-Level Fusion Network.

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Cardiovascular disease (CVD) poses a significant global health threat.
• Electrocardiogram (ECG) is a crucial diagnostic tool for CVD, particularly for detecting arrhythmias.
• Deep learning methods, especially those incorporating attention mechanisms, have shown promise in ECG analysis, but often focus on time-point weighting.

Purpose of the Study:

• To introduce a novel deep learning model, the Beat-Level Fusion Net (BLF-Net), for multiclass arrhythmia classification.
• To address the limitation of existing methods by applying attention mechanisms at the heartbeat level rather than just time points.
• To improve the accuracy and interpretability of automated arrhythmia detection from ECG signals.

Main Methods:

• The proposed BLF-Net segments long ECG signals into individual heartbeats.
• A neural network extracts features from each heartbeat.
• An attention mechanism assigns weights to heartbeat features based on their diagnostic contribution.

Main Results:

• The BLF-Net demonstrated superior performance compared to state-of-the-art methods on six classification tasks using the PTB-XL database.
• Visualization of the attention mechanism weights provided insights into the model's decision-making process.
• The model achieved high accuracy in multiclass arrhythmia classification.

Conclusions:

• The BLF-Net offers an effective and automated approach for arrhythmia classification.
• Applying attention at the heartbeat level enhances the diagnostic capability of ECG analysis.
• This method has the potential to significantly aid cardiologists in diagnosing arrhythmias.