Research On Denoising And Super-resolution Algorithms For Medical CT Images Based On Convolutional Neural Network

With the development of CT(Computed Tomography)technology,CT imaging has been widely used in the medical field for the detection of various diseases.However,the quality of the image may be degraded for various reasons in the process of CT imaging.Therefore,denoising and super-resolution processing of CT images are of great significance.This paper firstly reviews the history and basic principles of CT imaging technology,analyses the main noise sources and the main factors that restrict the imaging resolution in the process of CT imaging,and discusses the existing methods of CT image denoising and super-resolution.Then,the basic principles and applications of convolutional neural network are introduced.Because convolutional neural networks have powerful feature learning and mapping capabilities,they have certain advantages over traditional methods in image restoration tasks such as image denoising and super-resolution.Therefore,this paper intends to implement a denoising and super-resolution algorithm for CT images based on convolutional neural networks,respectively.Aiming at the complex noise existing in CT images,this paper proposes a convolutional neural network for CT image denoising,and the proposed network mainly improves the denoising performance through residual learning,batch normalization and dense connection.Both residual learning and batch normalization can improve the training efficiency of the network,and the combination of the two can fully utilize their respective properties to achieve complementary advantages so as to further improve the denoising performance of the network.Besides,the simplified dense connection mechanism effectively reduces the computational cost of the network while maintaining the reuse of feature maps at each layer.The experimental results show that the proposed network is advanced in the comprehensive denoising performance.For the problem of how to improve the resolution of CT images,this paper proposes a convolutional neural network for CT image super-resolution.Since batch normalization leads to performance degradation in super-resolution tasks,the proposed network removes all batch normalization layers while retaining the residual learning and dense connection mechanism.In addition,sub-pixel convolutional layer is introduced to enlarge the image size,which not only improves the super-resolution performance,but also reduces the computational cost.The experimental results show that the proposed network can effectively restore the image details while enlarging the image size.