<p>Magnetocardiography (MCG) provides a non-invasive method to record the magnetic fields produced by the heart’s electrical activity, which is essential for diagnosing cardiac conditions. However, the analysis of MCG signals is challenging due to their high dimensionality and complex geometric structure. This study aims to develop an advanced method for clustering and classifying MCG signals by integrating transformer networks with Riemannian manifold analysis, which leverages the strengths of both approaches to enhance signal analysis. We propose a novel pipeline consisting of three main modules: preprocessing and feature extraction, where MCG signals are prepared and initial features are extracted; transformer network processing, where a transformer architecture is used to learn higher-level representations from the embedded features; and Riemannian manifold analysis and clustering/classification, where the learned representations are analyzed on a Riemannian manifold to perform clustering and classification tasks. Our method demonstrated superior performance in clustering and classifying MCG signals compared to traditional methods. The transformer network effectively captured complex patterns in the signals, while the Riemannian manifold analysis preserved the geometric structure, leading to more accurate and interpretable results. The integration of transformer networks and Riemannian manifold analysis provides a powerful framework for MCG signal analysis, with the potential to improve the diagnosis of cardiac conditions and could be extended to other areas of medical signal analysis.</p>