Integrating ECG and PCG Signals through a Dual-Modal ViT for Coronary Artery Disease Detection

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cardiology

Background:

• Cardiovascular disease (CVD), particularly coronary artery disease (CAD), is a major global health concern and leading cause of mortality.
• Current CAD screening methods often lack accuracy, non-invasiveness, or cost-effectiveness, highlighting the need for improved diagnostic tools.

Purpose of the Study:

• To develop and evaluate a novel, non-invasive classification framework for accurate coronary artery disease (CAD) detection.
• To integrate electrocardiogram (ECG) and phonocardiogram (PCG) signals using an advanced AI model for enhanced diagnostic performance.

Main Methods:

• A Vision Transformer-based framework, Co-Attention Dual-Modal ViT (CAD-ViT), was designed to process both ECG and PCG signals.
• Key innovations include a Co-Attention mechanism for cross-modal feature interaction and a Dynamic Weighted Fusion (DWF) module for adaptive signal integration.
• The model was validated on a private clinical dataset and two public datasets.

Main Results:

• CAD-ViT achieved high performance on a private dataset, with an accuracy of 97.08%, precision of 97.18%, specificity of 98.52%, F1-score of 97.04, and recall of 96.94%.
• Validation on public datasets confirmed the model's robustness and generalization capabilities in CAD detection.
• The co-attention and dynamic fusion mechanisms effectively leveraged complementary information from ECG and PCG signals.

Conclusions:

• The proposed CAD-ViT framework demonstrates significant potential for accurate and non-invasive screening of coronary artery disease (CAD).
• Multimodal signal integration using advanced AI techniques offers a promising avenue for improving cardiovascular disease diagnostics.
• The model's high performance suggests feasibility for practical clinical deployment in CAD screening.