Detection of Atrial Fibrillation Using 1D Convolutional Neural Network

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cardiology

Background:

• Atrial fibrillation (AF) detection is vital due to its link with embolic stroke risk.
• Current AF detection methods often use complex 2D spectrograms, leading to high computational costs.
• There is a need for more efficient and accurate AF detection systems.

Purpose of the Study:

• To develop a simplified and accurate AF detection method using a 1D convolutional neural network (CNN).
• To reduce the computational complexity and implementation cost of AF detection systems.
• To improve the detection accuracy of atrial fibrillation from ECG signals.

Main Methods:

• An end-to-end 1D CNN architecture was designed for AF detection.
• The impact of convolutional block components on accuracy was investigated.
• Grid search was employed for optimal CNN hyperparameter tuning.
• A length normalization algorithm was developed for variable-length ECG records.

Main Results:

• The proposed 1D CNN achieved an average F1 score of 78.2%.
• The method demonstrated improved detection accuracy compared to existing deep learning techniques.
• The 1D CNN exhibited lower network complexity.

Conclusions:

• The 1D CNN approach offers a simple yet effective method for AF detection.
• This method provides a balance between high detection accuracy and reduced computational complexity.
• The findings suggest a promising direction for efficient AF screening and management.