Sparse view computed tomography (CT) is an effective way to lower the radiation exposure, but results in streakingartifacts in the constructed CT image due to insufficient projection views. Several approaches have been reported for fullview sinogram synthesis by interpolating the missing data into the sparse-view sinogram. However, current interpolationmethods tend to generate over-smoothed sinogram, which could not preserve the sharpness of the image. Such sharpnessis often referred to the region boundaries or tissue texture and of high importance as clinical indicators. To address thisissue, this paper aims to propose an efficient sharpness-preserve spare-view CT sinogram synthesis method based onconvolutional neural network (CNN). The sharpness preserving is stressed by the zero-order and first-order differencebased loss function in the model. This study takes advantage of the residual design to overcome the problem of degradationfor our deep network (20 layers), which is capable of extracting high level information and dealing with large sampledimensions (672 x 672). The proposed model design and loss function achieved a better performance in both quantitativeand qualitative evaluation comparing to current state-of-the-art works. This study also performs ablation test on the effectof different designs and researches on hyper-parameter settings in the loss function.
展开▼