Research On Low-Dose CT Image Denoising Based On Deep Learning

The ionizing radiation of computed tomography(CT)scans may cause genetic mutations and increase the risk of cancer for patients.Low-dose CT(LDCT)scan can greatly reduce the threat of X-ray radiation to human health.Therefore,it has a promising prospect in the field of medical imaging.However,when the dose is reduced,the image quality is also significantly reduced.There is a certain degree of noise and artifacts in the reconstructed image and the characteristics of lesions may be covered by noise or artifacts,which will affect the clinical diagnosis.Therefore,reducing the radiation dose of CT and improving the quality of reconstructed image have always been a research hotspot in the field of CT,which has important scientific research and clinical value.Image post-processing method denoises the reconstructed LDCT image directly,which does not rely on the original data,which is extremely beneficial to the privacy protection of patients.In this paper,an image post-processing method based on deep learning is used to construct a denoising network model,which can reduce the noise level of LDCT while preserving the image structure details.The specific research contents are as follows:(1)LDCT denoising based on attention mechanism.In this study,the multi-dose level chest and abdominal CT datasets are used to improve the robustness of the network.Residual attention modules(RAM)are incorporated into the residual encoder-decoder convolutional neural network(RED-CNN)and generative adversarial network with Wasserstein distance(WGAN)to learn features that are beneficial to improving the performances of denoising networks,and developed models are denoted as RED-CNN-RAM and WGAN-RAM,respectively.In detail,RAM is composed of a multi-scale convolution module and an attention module built on the residual network architecture.The residual network architecture solves the problem of network degradation with increased network depth.The function of the attention module is to learn which features are beneficial to reduce the noise level of low-dose CT images to reduce the loss of detail in the final denoising images.(2)LDCT denoising based on hybrid cascade network.In this study,the multi-dose level head CT datasets are used to train the network and construct a CT denoising network suitable for multi-dose level.The popular low-dose CT(LDCT)denoising network outputs denoised images through the end-to-end mapping between the LDCT image and its corresponding ground truth.The limitation of this method is that the reduced image noise level may not meet the doctor’s diagnostic needs.In view of this situation,we propose a hybrid cascade network to learn the feature of original LDCT and the outputs of the previous clone modules,which can be reused in each following clone module.The proposed network can achieve the goal of gradual denoising,which may be suitable for multi-noise levels denoising tasks.For the clone module,we designed a residual dual-branch convolution neural network(RDB-CNN).The residual learning is used to improve network performance,and the dual-branch structure increases the network width,thus the RDB-CNN is expected to extract as many feature information as possible.