Hybrid Modeling and Measurement-Augmented Neural ODE Framework for Arterial Waveform Reconstruction and Cardiovascular Risk Stratification
Academic Article in Scopus
Despite the increasing sophistication of cardiovascular modeling techniques, a significant gap persists between advanced hemodynamic simulations and clinically accessible measurements. In this work, we propose a measurement-augmented Neural Ordinary Differential Equation (Neural ODE) framework that reconstructs time-resolved arterial waveforms from a reduced set of non-invasive scalar clinical variables, including systolic and diastolic blood pressure (SBP/DBP), heart rate (HR), pulse wave velocity (PWV), age, and ascending aortic diameter. The clinically measurable scalar vector constitutes the primary measurement reference of the framework. The Pulse Wave Database (PWDB) is used exclusively as a controlled in-silico validation environment to evaluate reconstruction consistency under known parameter conditions. Arterial wave propagation is formulated as a continuous-time dynamical process using a Neural Ordinary Differential Equation approach, enabling the reconstruction of arterial pressure, flow velocity, and luminal area signals from non-invasive measurements. Physiologically meaningful parameters - such as heart rate, arterial compliance, peripheral resistance, and pulse wave velocity - are embedded within the model to constrain learning and improve robustness across individuals. The framework is trained on arterial waveform data and employs adjoint sensitivity methods to efficiently estimate model parameters. Additionally, the same input vector enables clustering-based vascular phenotype stratification. The proposed framework acts as a measurement augmentation tool, bridging clinically accessible scalar measurements and time-resolved arterial dynamics, with potential applications in personalized cardiovascular assessment.
