WarpPINN: Cine-MR Image Registration with Physics-Informed Neural Networks

Explore a cutting-edge approach to cardiac image analysis in this 58-minute talk by Francisco Sahli. Delve into the development of WarpPINN, a physics-informed neural network designed for image registration in cine magnetic resonance imaging. Learn how this innovative method quantifies local heart deformations, providing valuable insights for diagnosing cardiomyopathies. Discover how WarpPINN incorporates tissue incompressibility and hyperelastic behavior to generate accurate strain estimations. Examine the use of Fourier feature mappings to overcome neural network limitations and capture discontinuities in strain fields. Evaluate the algorithm's performance on synthetic examples and a benchmark dataset of healthy volunteers. Gain understanding of how WarpPINN outperforms existing methodologies in landmark tracking and delivers physiologically accurate strain measurements in radial and circumferential directions. Consider the potential applications of this technique for improved heart failure diagnosis and general image registration tasks.