Review of Sparse- View or Limited-Angle CT Reconstruction Based on Deep Learning

Computed tomography (CT) technology is widely used in clinical medical diagnosis thanks to the excellent visualization of the CT imaging technology for the internal cross-sectional structure of objects. Because X-ray radiation will be harmful to the human body, it is demanded to reduce the dose of X-ray radiation to patients by reducing the X-ray intensity of the scan or number of the view of the scan. However, low-dose CT images reconstructed from sub-sampling projection data will produce severe stripe artifacts and noise. In recent years, deep learning techniques have developed rapidly, and convolutional neural networks have shown great advantages in image representation and feature extraction, helping to achieve high speed and quality CT reconstruction from sparse-view or limited- angle projection data. This paper mainly focuses on the sparse-view or limited-angle CT reconstruction techniques, and reviews the latest research progresses of deep learning techniques in CT reconstruction in five directions, including image post- processing, sinogram domain pre-processing, joint processing of dual domain data, iterative reconstruction algorithms, and end-to- end mapping reconstruction. In the end, we analyze the technical characteristics, advantages, and limitations of existing sparse- view or limited-angle CT reconstruction methods based on deep learning, and discuss possible future research directions to address these challenges.