Reconstruction of Aortic Waveforms from Peripheral Data using Physics Informed Neural Networks

Aortic waveforms provide crucial hemodynamic information on cardiovascular health and disease states. However, access to aortic waveforms on an individualized basis non-invasively is limited. One opportunity is to use peripheral artery hemodynamics to reconstruct aortic waveforms. This reconstruction at personalized levels presents a significant challenge due to the complex hemodynamic interactions that alter pulse morphology. Traditional approaches to physiological modeling rely on either physics-based principles or purely data-driven methods, both of which have inherent limitations in personalization and scalability. In this work, we introduce a novel physics-informed neural network (PINN) framework that solves these gaps while leveraging peripheral artery data for accurate and personalized aortic waveform reconstruction. Our proposed framework estimates individualized hemodynamic parameters without requiring aortic measurements, enabling scalable and patient-specific cardiovascular modeling. We compare our PINN model against a conventional data-driven neural network model, where the normalized waveform RMSEs were 0.17 and 0.47, respectively, demonstrating a 64% improvement in error. Furthermore, our model extrapolates well to unseen locations along the aorta, addressing a key limitation with typical neural network models. These findings suggest that our PINN framework offers high-fidelity physiological flow and pressure modeling at the individual levels.