Multiscale dilated convolutional neural network for Atrial Fibrillation detection.

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence in Healthcare

Background:

• Atrial Fibrillation (AF) is a common heart arrhythmia increasing with age, raising stroke and mortality risks.
• Current automated AF monitoring methods struggle to balance accuracy with computational efficiency.
• There is a critical need for effective, low-cost, and efficient AF detection solutions.

Purpose of the Study:

• To introduce a novel framework, AF-MSDC, for precise and computationally efficient Atrial Fibrillation detection.
• To develop and evaluate Multi-Scale Dilated Convolution (MSDC) modules for multi-scale feature extraction from ECG data.
• To achieve an optimal balance between prediction accuracy and computational resource utilization.

Main Methods:

• Developed a framework named AF-MSDC, integrating three Multi-Scale Dilated Convolution (MSDC) modules.
• Trained and validated the AF-MSDC model on established datasets: MIT-BIH Atrial Fibrillation Database and Physionet Challenge 2017.
• Compared AF-MSDC performance against state-of-the-art (SOTA) methods in AF detection.

Main Results:

• The AF-MSDC model demonstrated superior performance compared to existing SOTA methods.
• Achieved high sensitivity (99.45%) and specificity (99.64%) on the MIT-BIH AFDB dataset.
• Attained an [Formula: see text] score of 85.63% on the Physionet Challenge 2017 AFDB dataset.
• The model utilizes only a quarter of the parameters of a ResNet, highlighting its computational efficiency.

Conclusions:

• The AF-MSDC framework provides a highly accurate and efficient solution for automated Atrial Fibrillation detection.
• Its low computational footprint makes it ideal for integration into edge computing frameworks and wearable ECG devices.
• This innovation holds significant potential for early AF diagnosis, improving patient outcomes and reducing healthcare burdens.