Metal Artifacts Removal In CT Based On Deep Learning

Computed tomography(CT)is an important technical means in medical diagnosis,but if the patient’s body contains metal implants such as artificial hip joints,spinal implants,etc.It will cause the CT to appear a contrast between light and dark in the CT image.The metal artifacts will cover up the human organs and tissues,thus affecting the doctor’s judgment of the patient’s condition.Therefore,it is very important to study the metal artifact removal(MAR)algorithm in CT images.Because traditional algorithms for MAR are not effective in clinical applications,either the calculation time is long or the artifact removal is incomplete.In recent years,deep learning has gradually been widely used in the field of computer vision,so this paper studies the application of deep learning to the removal of metal artifacts in CT images.This paper first studies the use of convolutional neural networks to remove metal artifacts in the image domain.Two methods are proposed.One is to directly use convolutional neural networks to remove artifacts in CT images,and the other is to use generative adversarial networks to simulate artifact distribution of metal artifacts,and then subtract the simulated metal artifacts from the CT image to remove the metal artifacts.Both methods have achieved better results than traditional methods.At the same time,the advantages and disadvantages of the two methods are analyzed,and the phenomenon that metal artifacts cannot be removed well when encountering large-sized metals in the experiment is discussed.It is concluded that when the metal artifacts increase in the image domain,the metal artifacts cannot be effectively removed.Secondly,this paper studies the use of convolutional neural networks in the sinogram domain to remove metal artifacts,analyzes the influence of different expressions of the sinogram on the artifact removal effect.It is concluded that sinogram represented by parallel beam projection as the network input have a better artifact removal effect,and discuss whether the Inpainting method in natural images can be applied to the problem of artifact removal in the signal domain.It is concluded that the Inpainting method in natural images can effectively solve the problem in the signal domain.Experiments have compared the difference between removing artifacts in the signal domain and removing artifacts in the image domain when encountering large-sized metals.It is concluded that the signal domain can better deal with stronger metal artifacts than the image domain.Finally,this paper studies the use of deep learning to remove metal artifacts in the signal domain and the image domain together.First,the difference between removing artifacts from the signal domain and the image domain alone is analyzed,and then this paper proposes an end-to-end approach.The network structure avoids the distortion of the anatomical structure in the final result and minimizes the occurrence of secondary artifacts.Through experimental comparison,it is proved that this method benefited from the removal of artifacts from the signal domain and the removal of artifacts from the image domain.At the same time,this article discusses the possibility of artifact removal for 3D CT.Finally,by comparing experiments with other metal artifact removal algorithms,it is concluded that the method proposed in this paper has the best effect in removing metal artifacts...