Deep residual-dense network based on bidirectional recurrent neural network for atrial fibrillation detection

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Atrial fibrillation (AF) poses significant risks, including stroke and cerebral infarction.
• Traditional deep learning models struggle with interference and generalization in AF detection.
• Accurate and early detection of AF is crucial for patient health outcomes.

Purpose of the Study:

• To introduce an innovative deep residual-dense network integrated with a bidirectional recurrent neural network (RNN) for enhanced atrial fibrillation detection.
• To simplify feature extraction and create an end-to-end deep learning model for AF identification.
• To leverage an attention mechanism for effective feature fusion and extraction of high-value information.

Main Methods:

• Development of a novel deep residual-dense network combined with a bidirectional RNN architecture.
• Implementation of an end-to-end deep learning model for automated atrial fibrillation detection.
• Utilization of an attention mechanism to fuse diverse features and pinpoint critical information.

Main Results:

• The proposed model achieved a high accuracy of 97.72% in atrial fibrillation detection.
• Sensitivity and specificity were reported at 93.09% and 98.71%, respectively.
• The model demonstrated superior performance compared to existing methods.

Conclusions:

• The deep residual-dense network with bidirectional RNN offers a powerful and effective approach for atrial fibrillation detection.
• This method simplifies the detection process and improves diagnostic accuracy.
• The findings suggest a promising advancement in leveraging AI for cardiovascular health monitoring.