MetaInv-Net: Meta Inversion Network for Sparse View CT Image Reconstruction

X-ray Computed Tomography (CT) is widely used in clinical applications such as diagnosis and image-guided interventions. In this paper, we propose a new deep learning based model for CT image reconstruction with the backbone network architecture built by unrolling an iterative algorithm. However, unlike existing strategy to include many data-adaptive components unrolled dynamics possible, find that it enough only learn parts where traditional designs mostly rely on intuitions experience. More specifically, initializer conjugate gradient (CG) algorithm involved one of subproblems model. Other components, priors hyperparameters, are kept original design. Since hypernetwork introduced inference initialization CG module, makes proposed certain meta-learning Therefore, shall call meta-inversion (MetaInv-Net). The MetaInv-Net can be designed much less trainable parameters while still preserves its superior performance than some state-of-the-art models imaging. simulated real data experiments, performs very well generalized beyond training setting, i.e., other scanning settings, noise levels, sets.