Physics-Informed Self-Supervised Generation of Digital Brain MRI Phantoms from Weighted Images Using Differentiable MRI Simulation

This paper targets biomedical imaging by generating digital brain MRI phantoms from T1-, T2-, and proton-density-weighted MR images. The proposed physics-informed, self-supervised framework predicts T1, T2, and PD maps, then uses an MRI signal model to reconstruct the input images, without requiring ground-truth parametric maps or anatomical segmentation. The source also describes three generative architectures, including VAE and GAN approaches.