Multiscale feature enhanced gating network for atrial fibrillation detection.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cardiology

Background:

• Atrial fibrillation (AF) is a major cause of stroke and heart failure.
• Current deep learning methods for AF diagnosis using ECG are limited by noise and redundant features.

Purpose of the Study:

• To propose a novel multiscale feature-enhanced gating network (MFEG Net) for improved AF diagnosis.
• To enhance the robustness and accuracy of AF detection in noisy ECG signals.

Main Methods:

• Developed MFEG Net integrating multiscale convolution, adaptive feature enhancement (FE), and dynamic temporal processing.
• The FE module utilizes soft-thresholding, dilated convolution, and SE modules to refine features.
• The dynamic temporal module captures time-dependent patterns crucial for AF recognition.

Main Results:

• MFEG Net achieved 93.0% accuracy and 0.883 F1 score on the PhysioNet Challenge 2017 dataset.
• Demonstrated remarkable resilience to noise interference.
• Achieved high accuracies (0.908, 0.938) on independent CPSC2018 and AFDB datasets.

Conclusions:

• MFEG Net shows excellent performance and robustness for automatic atrial fibrillation detection from noisy ECGs.
• This deep learning approach represents significant progress over state-of-the-art methods.
• Potential to reduce the clinical burden of manual AF diagnosis.