Denoising Method Of Low-dose CT Images Based On Deep Learning

Computed tomography is an axial scanning technology,and it is one of the most widely used non-invasive imaging techniques in diagnosis and disease detection.CT scanning uses X-ray to scan a certain part of the patient repeatedly,and then takes the data obtained by the detector as the input for the computer to reconstruct the image,which has the advantages of high speed,high accuracy and short scanning time.However,in the process of CT scanning,high-dose radiation will cause a certain degree of harm to the human body,and induce hereditary,cancer and other diseases,which is a problem of great concern to patients and researchers.The low-dose CT scanning technology will increase the noise and artifacts in the reconstructed image,and affect the diagnosis results of doctors.Therefore,it has become an important research direction in the field of medical imaging to obtain the image quality to meet the clinical needs when reducing the radiation dose.The main contributions of this paper are as follows:(1)In the methods based on traditional deep network,the status of feature channels in the same layer is equal,which affects the extraction of information.In order to solve the problem,an attention mechanism U-Net(DAU-Net)for low-dose CT image denoising is proposed.In order to improve the ability of network to recognize noise and image content,an attention mechanism is introduced in the network.This attention mechanism is composed of channel attention module and pixel attention module.Through the organic fusion of different feature information captured by the two attention modules,a better feature representation can be obtained.In addition,the network can effectively combine shallow and deep features to avoid the loss of details.Experiments show that the network has a good effect on the removal of artifacts and noise,and further improves the noise removal of low-dose CT.(2)A multi-scale attention residual U-Net(MARU-Net)is proposed to remove artifacts and noises from low-dose CT images.Multi scale feature edcoding block and attention mechanism are introduced in the network to sense more relevant information and improve the performance.Firstly,in order to obtain more abundant feature information,multi-scale feature coding blocks are introduced into the encoder to extract artifacts of different scales,and these fused features are cascaded to the decoder to avoid information loss.Secondly,the pixel attention mechanism is introduced to enhance the feature extraction ability of the network to improve the network performance.The image quality is evaluated by subjective analysis and objective evaluation index scores,which proves that the proposed denoising method can significantly improve the image quality.It can obtain better noise suppression effect while maintaining the texture and edge information of low-dose CT image.