Physics-Informed Neural Networks Enhance Myocardial Perfusion MRI Quantification

A new arXiv preprint (2608.11282) proposes an extension to physics-informed neural networks (PINNs) for quantifying myocardial perfusion from cardiac magnetic resonance (CMR) imaging. The method integrates spatiotemporal implicit neural representations (INRs) to model the MR signal as a continuous function, improving accuracy, smoothness, and physical consistency. The approach addresses challenges in fitting tracer-kinetic models to dynamic contrast-enhanced data, which is sensitive to noise and acquisition variability. The study demonstrates promising results on realistic simulated CMR datasets, suggesting a potential advancement over conventional non-linear least squares fitting and prior PINN frameworks.