SS-SWT and SI-CNN: An Atrial Fibrillation Detection Framework for Time-Frequency ECG Signal.

Area of Science:

• Cardiology
• Biomedical Engineering
• Signal Processing

Background:

• Atrial fibrillation (AF) is a prevalent arrhythmia linked to severe health outcomes, including stroke and heart failure.
• Rapid and accurate detection of AF is crucial for mitigating patient morbidity and mortality.
• Current screening methods face challenges in efficiency and speed.

Purpose of the Study:

• To propose and evaluate a novel framework for detecting atrial fibrillation from time-frequency electrocardiogram (ECG) signals.
• To enhance the speed and accuracy of AF screening through advanced signal processing and machine learning.

Main Methods:

• Utilized specific-scale stationary wavelet transform (SS-SWT) to decompose 5-second ECG segments into 8 scales, selecting key time-frequency features.
• Developed a scale-independent convolutional neural network (SI-CNN) to process the selected ECG features as a 2D matrix.
• Designed a specialized convolution kernel within SI-CNN to capture ECG's time-frequency characteristics while preserving scale independence.

Main Results:

• Achieved high performance metrics on the MIT-BIH AFDB dataset, including 99.03% sensitivity, 99.35% specificity, and 99.23% overall accuracy.
• Demonstrated that the SS-SWT and SI-CNN framework effectively extracts ECG features, reduces redundancy, and accurately identifies AF signals.
• The proposed method simplifies feature extraction compared to traditional wavelet transforms.

Conclusions:

• The SS-SWT and SI-CNN framework provides an effective and accurate method for atrial fibrillation detection.
• The proposed approach shows significant potential for clinical application in rapid and efficient AF screening.
• This method addresses limitations in current AF screening by optimizing feature extraction and utilizing advanced deep learning techniques.