Low Dose CT Image Reconstruction Based On Deep Learning

Computed tomography(CT)takes advantage of differences in absorption of X-ray by different organs,measures the remaining number of photons after passing through the human body,and utilizes mathematical principles such as Radon transform to reconstruct CT images which contains information about internal organs of the human body.CT imaging technology can obtain the structure information of organs without damage,so it has a wide range of applications in clinical diagnosis and treatment.However,high dose Xray scanning can damage the organ structure.Decreasing tube current is the most practical way to lower radiation dose in clinic,but this will produce more noise.Besides,due to the limitations of equipment conditions and reconstruction algorithms,blur and artifacts also can occur in low dose CT reconstruction.These interference factors will seriously reduce the quality of the reconstructed image,which will affect the subsequent diagnosis and treatment.Therefore,low dose CT reconstruction is of great practical significance.In recent years,deep learning has achieved rapid development.Not only has it made great progress in high-level computer vision tasks such as image classification and segmentation,but also it has achieved great success in low-level image processing tasks such as image denoising and restoration.The main task of the article is to solve two kinds of problems in low dose CT reconstruction with the help of deep learning.(1)Blur problem in low dose CT reconstructed image.In this paper,the deblur problem in CT reconstruction was mathematically modeled,and was solved by using a combination of mathematical derivation and deep learning.The CT deblurring algorithm based on deep learning can not only deblur,but also suppress noise to preserve the image structure and edge information.(2)Artifacts in low dose CT reconstruction.The traditional analytic CT reconstruction method can quickly reconstruct CT images,but it cannot eliminate artifacts and noise.The paper proposed the residual FBP algorithm combined with deep learning and FBP algorithm,which can not only retain the speed advantages of traditional FBP algorithm,but also effectively suppress noise and remove artifacts.