UniCardio is an integrated generation framework that reconstructs low-quality cardiovascular signals (PPG, ECG, BP) and generates unrecorded signals using a multi-mode diffusion transformer. Its specialized model architecture handles various signal modes and integrates various mode combinations through a continuous learning paradigm. By leveraging the complementary characteristics of cardiovascular signals, it outperforms existing task-specific baseline models in signal removal, interpolation, and transformation tasks. The generated signals perform similarly to real signals in detecting abnormal health conditions and estimating vital signs, even in uncharted territory, ensuring expert interpretability.
