Title: Calibrate Everything: Self-Optimizing MRl Reconstruction via Joint Prior Adaptation

Abstract: Magnetic Resonance Imaging (MRI) reconstruction from undersampled k-space data is an ill-posed inverse problem often hindered by unknown or varying forward models across scans. Traditional approaches rely on fixed priors, such as precomputed coil sensitivity maps. We propose a self-calibrating reconstruction method based on a bilevel Plug-and-Play Deep Equilibrium (DEQ) model, which dynamically adapts both image and acquisition priors within a unified optimization. Unlike conventional approaches, our method requires no precomputed sensitivity maps or fixed sampling patterns. Experimental results demonstrate that our adaptive DEQ framework consistently outperforms state-of-the-art methods, achieving higher PSNR and SSIM across diverse imaging settings. This highlights its potential as a fully adaptive, data-driven solution, offering significant clinical utility for accelerated MRI reconstruction.