Research On Post-processing Algorithm Of Low-dose CT Denoising Based On Convolution Neural Network

Computed tomography technology has been widely used in the field of clinical screening and disease tracking because of its clear imaging and fast scanning speed.The emergence of computed tomography technology provides an effective reference for doctors to diagnose the disease,but recent studies show that the radiation generated by CT shooting will bring great hidden danger to the health of patients.Therefore,people pay more attention to the harm of Xray in CT technology.Researchers try to reduce the radiation dose to reduce the harm to the patient’s body,but the results show that reducing the radiation dose directly affects the quality of the image,because it will cause noise and artifacts in the acquired image,which is not conducive to the doctor’s diagnosis of the disease,and even cause misdiagnosis.Aiming at the problems of low-dose CT image,this thesis proposes a low-dose denoising method based on deep learning CNN,which improves the quality of low-dose CT image.The main work and content are as follows:(1)In order to improve the quality of CT image,this thesis proposes an unsupervised CT denoising network based on CNN.Due to the inherent harmfulness of X-ray and the influence of various tissue parts of human body,it is very difficult to obtain aligned data.At present,most of the methods are based on supervised training and need to use aligned data for training in order to play a better denoising level,but the performance on non-aligned data is unsatisfactory.Therefore,in order to solve the problem that the non-aligned data cannot be trained with the supervised method and get better results,this thesis makes corresponding adjustments and improvements to the original generator,discriminator and loss function based on Cycle GAN.The final results show that the proposed method has good performance on low dose CT images.(2)Considering that the network structure of most method is more and more complex and the scale is larger and larger,the waste of resources and the decline of efficiency are followed.This thesis designs and improves a lightweight network low-dose denoising method based on deep learning.The reliable and effective lightweight denoising network is selected,and then the loss function of convolutional neural network is fully used in the guidance of network training process.The scientific and efficient loss function is configured,and the excellent lowdose CT denoising network is trained together with the lightweight denoising network.The results show that the proposed method can improve the efficiency while maintaining the quality requirements of low-dose CT images.