LAUSR

Helical computed tomography (CT) scans are often performed to obtain three-dimensional images of an object that is longer than the detector. However, the existing quasi-exact and exact reconstruction methods, such as re-binning and Katsevich algorithm, generate interpolation errors or require high computational power. In this work, we propose a method to reconstruct helical CT projections by iteratively reducing helical artifacts. In each iteration, a convolutional neural network (CNN)-based denoising technique is used to accurately segment the prior image (bone and soft tissue image). The results indicate that the proposed algorithm reduces helical artifacts to a significantly greater extent than the existing single slice re-binning (SSR) and weighted filtered backprojection (W-FBP) methods.