Deep learning for the classification of atrial fibrillation using wavelet transform-based visual images.

Area of Science:

• Cardiology
• Biomedical Engineering
• Signal Processing

Background:

• Atrial Fibrillation (AF) incidence is rising globally, increasing the need for effective diagnostic tools.
• Electrocardiogram (ECG) signal analysis is crucial for diagnosing AF.
• Morse Continuous Wavelet Transform (MsCWT) is an emerging feature extraction technique for ECG signals.

Purpose of the Study:

• To investigate the efficacy of MsCWT in classifying AF using ECG signals.
• To develop and evaluate a deep learning model for AF detection based on MsCWT-derived ECG images.

Main Methods:

• Utilized Morse Continuous Wavelet Transform (MsCWT) for feature extraction from ECG signals.
• Developed a deep learning machine to classify AF based on MsCWT-generated images.
• Trained, validated, and tested the deep learning model on ECG datasets.

Main Results:

• Achieved high average accuracies: 97.94% (training), 97.84% (validation), and 91.32% (test).
• Obtained excellent overall F1 scores: 97.13% (training), 96.86% (validation), and 89.41% (test).
• Demonstrated Area Under the Receiver Operating Characteristic Curve (AUC ROC) values exceeding 0.99 for training/validation and 0.9679 for the test set.

Conclusions:

• Deep learning models trained with MsCWT-based ECG images show superior performance for AF classification.
• The MsCWT technique significantly enhances diagnostic outcomes in ECG analysis.
• This approach holds promise for improving various signal-based diagnostic applications beyond ECG.