Deep arrhythmia classification based on SENet and lightweight context transform.

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Arrhythmia is a prevalent cardiovascular disease.
• Current computer-aided ECG analysis faces challenges with morphological variations in abnormal data.
• Effective arrhythmia identification is crucial for patient diagnosis and management.

Purpose of the Study:

• To propose a novel deep learning method for accurate ECG-based arrhythmia classification.
• To address the limitations of existing methods in handling diverse morphological changes in ECG signals.
• To develop a robust and efficient system for automated arrhythmia detection.

Main Methods:

• Feature extraction from ECG signals using Continuous Wavelet Transform (CWT).
• Development of a lightweight context transform block, enhanced with Squeeze-and-Excitation (SE) networks and linear transformation.
• Classification of arrhythmia types using the proposed deep learning architecture.

Main Results:

• The proposed method demonstrated high accuracy in classifying arrhythmias on the MIT-BIH arrhythmia database.
• The novel lightweight context transform block effectively captures relevant ECG features.
• Validation confirmed the method's efficacy in distinguishing various types of arrhythmias.

Conclusions:

• The proposed deep learning method offers a promising advancement for computer-aided arrhythmia detection.
• The integration of CWT and the enhanced transform block improves classification performance.
• This approach has the potential to enhance the clinical diagnosis of cardiac arrhythmias.