Deep Learning Based On Super-resolution Reconstruction Of Lung CT Images

At present,CT imaging technology is widely used in clinical disease screening.The X-ray radiation in CT images will be deposited in the human body,and when the deposited amount exceeds the human body tolerance dose,it will cause harm to the human body.Therefore,it is urgent to reduce the CT radiation dose.However,the reduction in radiation dose can lead to a decrease in the quality of CT imaging,which in turn affects the doctor’s diagnosis of the disease.In order to obtain clearer CT images at low doses,this paper applies the super-resolution reconstruction method to clinical lung CT images to improve the quality of lung CT images obtained at low doses.In recent years,using deep learning algorithms in super-resolution reconstruction techniques can achieve better results than interpolation and reconstruction algorithms.This paper focuses on improving the reconstruction quality and reconstruction speed of lung CT images,and studies the problems existing in the deep learning-based super-resolution reconstruction algorithm of lung CT images.This research mainly includes the following contents:Aiming at the problem of low definition of lung CT images after neural network reconstruction,a network model for super-resolution reconstruction of lung CT images based on DCA is proposed.The network uses the convolution layer of the residual channel attention module to focus on the feature information of low-resolution lung CT images,and the channel attention mechanism can extract the statistical information of the channel,and finally fuse the local and global information to improve the quality of the reconstructed lung CT images.Aiming at the problem of slow operation speed of neural network,a network model for super-resolution reconstruction of lung CT images based on SCRes Net is proposed.The skip connection residual network in the feature extraction module of the network can effectively extract low-frequency and high-frequency information in lung CT,and the sub-pixel convolution in the reconstruction module is connected in parallel with the reconstruction convolution module,which not only improves the performance of lung CT reconstruction clarity,but also improving the speed of the network.Aiming at the problem of single-scale training of neural networks,a network model for super-resolution reconstruction of lung CT images based on LPRN is proposed.The network parallels the original SRRes Net network,and removes the BN layer in the SRRes Net residual module.The network parallels the original SRRes Net network,and removes the BN layer in the SRRes Net residual module.Through the dense series connection between the residual blocks,the feature information of the images between the residual blocks is deeply fused to enhance the network to perception of image features.To sum up,this paper selects the lung,lung apex,and aorta from the lung cancer CT image data of the American Cancer Genome Atlas(TCGA)for experiments.The experiments show that the three reconstruction algorithms be proposed in this paper respectively solve the problems of low quality of lung CT image reconstruction,slow network operation speed and single-scale training.The research has a certain role in promoting the imaging of lung CT images.