Detection of Atrial Fibrillation From Variable-Duration ECG Signal Based on Time-Adaptive Densely Network and Feature...

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Signal Processing

Background:

• Atrial fibrillation (AF) is a prevalent arrhythmia with significant health risks.
• Current deep learning models, like CNNs, struggle with variable-duration ECG data.
• Accurate AF detection is crucial for patient outcomes.

Purpose of the Study:

• To develop an intelligent auxiliary diagnostic model for AF using ECG.
• To address the limitations of standard CNNs in handling variable-length ECG signals.
• To propose a novel, time-adaptive deep learning network for robust AF recognition.

Main Methods:

• Introduction of MP-DLNet-F, a novel time-adaptive densely network.
• Utilizing an MP-DLNet module to ensure compatibility with variable-duration ECG and 1D-CNN.
• Incorporating feature enhancement and data imbalance processing modules.
• Employing transfer learning for testing on heterogeneous datasets.

Main Results:

• MP-DLNet-F achieved 87.98% accuracy and 0.847 F1-score on the CinC2017 ECG database.
• Significant improvements of 21.81% in accuracy and 16.14% in F1-score on the CPSC2018 12-lead dataset via transfer learning.
• Demonstrated precise AF forecasting across different ECG durations and lead configurations.

Conclusions:

• MP-DLNet-F effectively overcomes the limitations of standard CNNs for variable-duration ECG analysis.
• The model shows strong generalization capabilities and adaptability to diverse medical signal processing tasks.
• This approach offers a promising solution for intelligent auxiliary diagnosis of AF and other medical conditions.