From Spins to Images to Digital Twins and Back Again
This lecture will reflect on the evolution of cardiovascular magnetic resonance (CMR) imaging from accelerated image reconstruction toward physics-informed digital twinning, highlighting how recent advances in artificial intelligence, computational modeling, and data-driven inference are reshaping the field. Beginning with the transition from supervised to self-supervised and generalizable reconstruction methods, the lecture will discuss how modern learning-based approaches can recover high-quality images from highly undersampled data while reducing dependence on fully sampled references. The presentation will further examine the growing role of biophysical modeling and synthetic data generation in overcoming one of the central challenges in medical AI: the lack of reliable ground truth. By integrating computational physiology, virtual MRI scanners, and controllable image synthesis, emerging frameworks now enable the generation of realistic paired datasets for training, validation, and uncertainty quantification of AI/ML methods in cardiovascular imaging. Examples including FlowMRI-Net, synthetic 4D/5D flow MRI, and physics-informed late gadolinium enhancement synthesis will illustrate how these tools are advancing robust image analysis and inference. Finally, the lecture will place these developments into the broader context of personalized cardiovascular medicine through the concept of digital twins. The SwissHeart initiative will serve as an example of how multimodal imaging, physiological measurements, biomarkers, and computational models can be integrated to create population-scale digital twins capable of characterizing healthy variation, tracking disease progression, and ultimately informing individualized prediction and intervention. The keynote will conclude with a perspective on future opportunities and challenges for uncertainty-aware, interpretable, and clinically deployable AI in cardiovascular imaging.
Read more
