Cardio-mmFlow: A Gaussian-Prior-Free Physics-Informed Flow Matching Framework for Electrocardiogram to mmWave Radar Synthesis

Continuous ECG monitoring is clinically valuable, but scaling it beyond electrodes to comfortable long-term use motivates contactless mmWave sensing. In practice, mmWave-to-ECG reconstruction is severely constrained by the scarcity of high-quality synchronized recordings. Therefore, we propose \textbf{Cardio-mmFlow}, a Gaussian-prior-free physics-informed flow matching framework that synthesizes mmWave radar signals from clinical ECG. It learns a direct transport trajectory between the latent manifolds of ECG and radar. Considering subject-dependent propagation differences, we incorporate a simplified mass--spring--damper inspired modulation and inject it into the flow dynamics via feature-wise linear modulation for personalization. Extensive experiments show that our system generates high fidelity radar data in both signal and latent domains. It significantly improves zero-shot downstream mmWave to ECG task, and enable Atrial Fibrillation classification with synthetic data. Further analyses evaluate the model interpretability.

Continuous electrocardiogram (ECG) monitoring is important for detecting and understanding heart conditions, but conventional electrode-based devices can be uncomfortable for long-term everyday use. Contactless sensing with millimeter-wave radar offers a promising alternative, as it can capture subtle chest movements related to cardiac activity without attaching sensors to the body. However, training reliable radar-based ECG models requires large amounts of synchronized radar and ECG data, which are difficult and costly to collect. This work proposes Cardio-mmFlow, a framework that generates realistic millimeter-wave radar signals from existing clinical ECG recordings. Instead of relying on random noise as the starting point, the method learns how to transform ECG representations directly into radar representations. It also incorporates a simplified physical model to account for differences across individuals. Experiments show that the generated radar signals are realistic, improve radar-to-ECG reconstruction without target-user training data, and support downstream tasks such as Atrial Fibrillation classification.