MCAF-Net: A non-invasive early screening method for coronary artery disease based on a multi-scale cross-modal model

Coronary artery disease (CAD) is one of the leading causes of death worldwide. Achieving early, non-invasive, and highly accurate detection of CAD is crucial for reducing its mortality rate. This study utilizes a high-sensitivity MEMS-based PCG-ECG Synchronous Auscultation System to construct a high-fidelity clinical dataset of synchronous PCG-ECG data for CAD. Leveraging deep learning technology, we developed a Multi-scale Cross-modal Attention Fusion Network (MCAF-Net). This network extracts multi-resolution features from synchronous PCG-ECG spectrograms using an improved residual network and performs feature interaction and fusion through Mutual Cross Attention (MCA), achieving high-precision detection of CAD (96.29% on the clinical dataset, 97.69% on the public dataset).