PUERT: Probabilistic Under-Sampling and Explicable Reconstruction Network for CS-MRI

Compressed Sensing MRI (CS-MRI) aims at reconstructing de-aliased images from sub-Nyquist sampling k-space data to accelerate MR Imaging, thus presenting two basic issues, i.e., where sample and how reconstruct. To deal with both problems simultaneously, we propose a novel end-to-end Probabilistic Under-sampling Explicable Reconstruction neTwork, dubbed PUERT, jointly optimize the pattern reconstruction network. Instead of learning deterministic mask, proposed subnet explores an optimal probabilistic sub-sampling pattern, which describes independent Bernoulli random variables each possible point, retaining robustness stochastics for more reliable CS reconstruction. A dynamic gradient estimation strategy is further introduced gradually approximate binarization function in backward propagation, efficiently preserves information improves quality. Moreover, our subnet, adopt model-based network design scheme high efficiency interpretability, shown assist exploitation subnet. Extensive experiments on widely used datasets demonstrate that PUERT not only achieves state-of-the-art results terms quantitative metrics visual quality but also yields model are customized training data.