Research On Metal Artefact Reduction Algorithm Of CT Image Based On Conditional Generative Adversarial Network

Computed Tomography(CT)is widely used to scan human tissues,but the presence of metal implants and other objects with a high attenuation factor causes metal artifacts to form in CT images.These low-quality CT images can have a negative impact on medical diagnosis and treatment evaluation.Therefore,it is necessary to metal artifact reduction(MAR)to improve the quality of CT images.To solve the problem of MAR,a MAR algorithm based on supervised conditional generation adversarial network is proposed.In order to provide matched training data with and without metal artifacts for the network,based on the idea of generating metal artifact images by simulating multi-energy projection of metal artifact images and metal prosthesis images,the fuzziness in the synthesized metal artifact images was removed by difference method,and the authenticity of the simulated images was improved.In order to enhance the generalization ability of the data,the preprocessing operation is carried out to enhance the data during the training.The network structure is designed based on the conditionally generated bottleneck network.The generator in the network is based on U-Net,and uses the second convolution in the lower and upper samples to replace the congestion.The residual structure of bottleneck type can compress the model parameters while thickening the network and stabilizing the network training.The discriminant in the network uses PatchGAN to capture high-frequency information in the image and achieve the purpose of generating metal-artifact removal images with higher clarity and detail.Meanwhile,cGAN loss and L1 loss are used as loss functions to constrain the training of the network model.The experimental results show that the proposed algorithm can improve the effectiveness of metal artifacts removal by using the combined image and clinical image based on peak signal-to-noise ratio and structure similarity.Compared with other MAR algorithms such as CNNMAR and U-Net,it has better artifact removal effect.