A hybrid model combining 1D-CNN and BERT for intelligent ECG arrhythmia classification

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Signal Processing

Background:

• Arrhythmia diagnosis is critical for preventing severe cardiac events.
• Manual electrocardiogram (ECG) interpretation is inefficient and lacks accuracy.
• Existing automated methods struggle with precise arrhythmia classification.

Purpose of the Study:

• To develop a novel deep learning model for accurate arrhythmia classification.
• To integrate the strengths of 1D-CNNs and BERT for enhanced ECG analysis.
• To improve the efficiency and accuracy of automated arrhythmia recognition.

Main Methods:

• Proposed ECGBert model integrating 1D-CNN and Bidirectional Encoder Representations from Transformers (BERT).
• Utilized signal preprocessing, segment encoding, and sequential feature extraction.
• Trained and evaluated the model on the MIT-BIH Arrhythmia Database.

Main Results:

• ECGBert significantly outperformed traditional methods and existing hybrid architectures.
• The model demonstrated superior performance across multiple evaluation metrics.
• ECGBert effectively captured long-range dependencies in ECG signals.

Conclusions:

• ECGBert offers a robust and generalizable approach for intelligent ECG analysis.
• The model provides a new methodological framework for deep learning in medical signal processing.
• This work advances automated arrhythmia recognition and cardiovascular disease diagnosis.