NPB-REC: A non-parametric Bayesian deep-learning approach for undersampled MRI reconstruction with uncertainty estimation

Samah Khawaled, Moti Freiman

Abstract: Highlights•NPB-REC: A non-parametric Bayesian deep-learning method for reconstruction of MRI images from under-sampled k-space data.•Full characterization of the posterior distributions of the reconstructed images.•Uncertainty estimate of the reconstructed images that correlates with the reconstruction error, anatomical distribution shift and undersampling mask distribution shift.•Enables a safer utilization of deep-learning-based reconstruction models in the clinical setting.