Research On Deep Learning Based CT Image Quality Enhancement Algorithm

With the development of modern medical technology,low dose Computed Tomography(CT)has attracted more and more attention due to its low radiation dose.However,low-dose artifacts are easy to appear in low-dose CT,and there are many interferences in the process of image imaging and transmission.As a result,the quality of CT images is degraded,which is not conducive to the diagnosis and treatment of the disease by medical staff,nor to the subsequent segmentation,feature extraction and classification.Therefore,the quality enhancement of CT images is increasingly important.Aiming at the problems existing in the medical image quality degradation,this thesis proposes a CT image quality enhancement method combined with the image noise level estimation to handle de-noise and super-resolution at the same time.This method can effectively remove the low-dose artifacts and noise in the medical image,and improve the resolution,so as to enhance the quality of the medical image and improve the visual effect of the image.To solve the problem that it is difficult to estimate the noise level in CT images,an image noise level estimation network combining Inception structure and dense connection structure is proposed to estimate the noise level in images.Inception architecture is used to extract relevant noise characteristics in multiple sensory fields,while dense connection architecture supports feature reuse,ensuring the transmission of features between networks and avoiding the problem that noise features weaken or disappear due to the increase of network depth.To solve the problem of multiple degradation in CT images,an improved residual dense connection network is proposed to de-noise and super-resolve low-quality CT images.The residual structure features can increase the depth of the network,and the dense connection mode can increase the width and feature dimension of the network,so as to increase the non-linear ability of the whole network to better represent the nonlinear mapping relationship between input and output.For the problem that there are many texture details in CT images,the joint loss function which combines perceptual loss and pixel level loss is used to constrain the quality enhancement network,so that the reconstruction performance of enhanced image edges and texture details is better.The experimental results on the public medical data set TCIA show that the noise level estimation method proposed in this thesis can estimate the noise level in the image with an accuracy of 99.5%,and the image quality enhancement method has a high peak signal-to-noise ratio(PSNR)and structural similarity(SSIM).Experimental results show that the image quality enhancement method proposed in this paper can effectively remove the noise in the medical image,improve the image resolution and enhance the image quality.