K-Means-Based Decision-Level Fusion of ECG and MCG for Cardiac Source Imaging

Magnetocardiography (MCG) and electrocardiography (ECG) are non-invasive techniques for mapping the heart’s electrical activity, but each faces challenges regarding spatial resolution and noise robustness. While fusing MCG and ECG data can enhance the localization accuracy of cardiac source imaging, conventional methods require synchronous signal acquisition or processing, which is often technically challenging and clinically impractical. This paper introduces a novel, synchronization-free decision-level fusion framework. The method first performs independent source imaging for each modality to obtain distributed power maps. Subsequently, K-means clustering is applied in a joint feature space, combining the spatial coordinates of cardiac sources with their reconstructed power from both MCG and ECG, to fuse the source imaging results. In simulations across various source directions, locations, and signal-to-noise ratios, the proposed fusion method demonstrated improved spatial resolution in source imaging compared to single-modality approaches. This work offers a new paradigm for multimodal cardiac source imaging.