Artifact Removal Using GAN Network For Limited-Angle CT Reconstruction

Computer tomography(CT)technology has become one of the greatest imaging technologies in today's society,and has deeply affected human life.The role of X-ray CT is to scan the internal structure information of the object to obtain the projection data to perform the reconstruction process.In the past few decades,this technology has been vigorously developed.The most widely used algorithm is the Filtered Back Projection(FBP)algorithm.When the projection data is complete,this method can reconstruct a high Quality images.For X-ray CT to obtain a clear imaging effect,the scanning angle range should exceed 180 °,but in some practical applications,it is limited by factors such as the scanning environment,the structure of the scanned object,and the amount of X-ray radiation.The object is scanned within the range of rotation angle(less than 180 °),and the projection data obtained at this time is incomplete.For example: due to(1)in the industrial field,the on-site inspection is restricted by the pipeline environment,and the X-ray source can only rotate within a limited rotation angle range,so the obtained projection data is incomplete.(2)In the medical field,when using dental CT to scan objects,due to the influence of the object's geometric structure of the scanning target,the data can only be scanned within a limited range of rotation angle;when using CT to image the chest cavity and breast and carry out medicine At diagnosis,the situation is similar.(3)In addition,in some applications,in order to save scanning time or reduce radiation dose,it is also a good measure to collect projection data within a limited rotation angle.In the above-mentioned finite angle CT reconstruction problem,if the FBP algorithm is used for reconstruction,the reconstructed image obtained will have obvious landslide artifacts.Therefore,the reconstruction of limited-angle CT images with high quality has important academic value and important commercial value,and is also a hot research direction in the current industry and academia.Based on deep learning,this paper reconstructs CT images with limited angles to obtain clear images.This method is based on the original WGAN(Wasserstein GAN),which transforms the network structure and incorporates an improved joint loss function to remove artifacts while preserving complete details and sharp edges.After experimental comparison,the reconstruction effect of this method is very close to that of complete projection data reconstruction image.Compared with several other traditional classical methods,the WGAN-based network proposed in this paper has better performance in terms of artifact reduction,feature retention and computational efficiency.