Research On Low-dose CT Image Reconstruction Algorithm Based On Neural Network

Computed Tomography(CT)can quickly Computed Tomography information about the internal structure of human body,which is helpful for doctors to diagnose and treat diseases.However,excessive X-ray radiation can increase the risk of patients with disease,so low-dose CT scan came into being.However,if the radiation dose is reduced,quantum noise will be introduced in the projection process,which will cause the quality of the reconstructed image to be degraded and the accuracy of the reconstructed image can not be guaranteed.The core of low-dose CT image reconstruction technology lies in the optimization and improvement of algorithm,which is the key factor to determine the speed and image quality of the whole reconstruction process.With the development of computer technology,researchers have found that deep neural network can extract high-level features from low-level features and apply it to low-dose CT reconstruction,which can effectively improve the quality of reconstructed images.Therefore,it is of great significance to study more efficient and accurate low-dose CT image reconstruction algorithm combined with depth learning for improving the efficiency and quality of medical and other related fields.In order to solve the problem of noise and artifacts in low-dose CT image reconstruction,a low-dose CT image reconstruction algorithm based on improved U-Net is proposed in this paper.The U-Net automatic encoder can suppress the noise in the process of layer-by-layer feature coding and realize the automatic denoising of image,at the same time,in order to increase multi-scale information in the upper sampling,a new network model is built after embedding jump connection,adjusting network structure and input-output.Experimental results show that the PSNR,SSIM and RMSE indices are improved by 21.699%,2.263% and40.833%,respectively.The model can improve the image quality and outperform other algorithms in image denoising.In order to solve the problem that the generated network is noisy,a dual-domain antagonism network based on U-Net residuals structure is proposed,and the perceived loss function is introduced,this directs the generator to generate CT images that contain more detailed information.The ability of the network to improve the quality of the real image is discussed,and the edge details are improved.The experimental results show that the network can suppress the noise,achieve more image detail.