Abstract

Reconstruction of 3D medical imaging data, including but not limited to Computed Tomography (CT), Magnetic Resonance Imaging (MRI), and Confocal Microscopy, has benefited from the advancement of computing technologies over the past decades. Rather than depending on the full views (projections) of the 2D source data for reconstruction, by leveraging the prior knowledge about the distribution of the projection and the 3D image, it becomes gradually feasible that only a limited (i.e., sparse) views will be needed to reconstruct the 3D image with similar quality. Such a feature can significantly reduce the scan time and required dosage (for CT) for imaging. This work will extensively review the technological progress of the sparse and ultra-sparse medical image reconstruction from the compressed sensing framework to the deep learning-based reconstruction.

© 2023 The Author(s)