Project Summary/Abstract Image reconstruction has been an integral part of MRI technology development over the past decades, spanning techniques such as parallel imaging, compressed sensing, low-rank matrix models, methods for novel encoding strategies/k-space trajectories, and more recently deep learning (DL) approaches. Our TRD in the first phase has been at the forefront of these developments, introducing multiple new technologies for improved DL reconstruction and training, interpretable image denoising, and fast iterative algorithms, which were applied to Cartesian, non-Cartesian and novel encoding strategies beyond Fourier encoding. These new MRI reconstruction methods have also pushed the imaging technologies in other TRDs, CPs and SPs forward to target higher resolutions and acceleration rates at lower signal-to-noise ratios (SNR), as well as to combine information across multiple nuclei or even modalities. Consequently, these targets necessitate newer technologies for image reconstruction and denoising, an interplay with image acquisition, and new approaches for multi-nuclei and multi-modal computational imaging. Each of these directions correspond to a specific inverse problem with its own distinct forward operator dictated by the underlying imaging physics. Our goal in this TRD is to link these inverse problems through the lens of intelligent physics-driven technologies that synergistically utilize imaging physics and advances in DL methods. For accelerated high- resolution imaging, which remains a focal point for our TRD, we will develop a new computational imaging pipeline that explicitly combines interpretable denoising methods and physics-driven DL reconstruction. These will be complemented by new technologies for improved denoising of MR image series, and for self-supervised training of physics-driven DL reconstruction from few examples without ground-truth data. On the acquisition side, we will concentrate on parallel transmit technology for UHF ima