MDD2DG-IRA: Multivariate Degree Distribution to Dynamic Graph With Inter-Channel Relevance Attention Mechanism for...

Area of Science:

• Cardiology
• Graph Theory
• Machine Learning

Background:

• Analyzing multi-channel Electrocardiogram (ECG) signals is crucial for diagnosing cardiac conditions.
• Traditional methods often rely on fixed features, limiting their ability to capture complex signal dynamics.

Purpose of the Study:

• To introduce a novel methodology (MDD2DG with IRA) for analyzing multi-channel ECG signals.
• To improve the accuracy and efficiency of cardiac disease detection using dynamic graph models.
• To explore signal connections across different ECG channels.

Main Methods:

• Multi-channel ECG signals were transformed into visual graphs to extract degree distribution features.
• Degree distributions were mapped into dynamic graphs using a neural network with an Inter-channel Relevance Attention (IRA) mechanism.
• Graph Convolutional Neural Networks (GCNNs) and a multilayer perceptron were used for feature extraction and classification, incorporating multi-scale position embedding for efficiency.

Main Results:

• Achieved 99.94% classification accuracy in distinguishing myocardial infarction (MI) subtypes and healthy controls (HC).
• The MDD2DG approach demonstrated superior recognition accuracy compared to traditional complex network methods.
• The multi-scale position embedding significantly enhanced model processing efficiency.

Conclusions:

• The MDD2DG with IRA methodology offers a robust approach for analyzing complex physiological signals like multi-channel ECG.
• This method provides a promising tool for improving clinical diagnosis and early detection of cardiac diseases.
• The dynamic graph modeling approach overcomes limitations of fixed feature extraction in traditional methods.