Low Dose CT Image Enhancement Based On Generative Adversarial Network

Computerised Tomography(CT)is a commonly used medical imaging technique that uses X-rays to project in all directions of the measured object and reconstructs an image of the attenuation coefficient of each part of the object from the projection data.The technique is widely used in clinical diagnosis and disease treatment as it can clearly visualise structures,lesions and tissues in certain parts of the body.However,patients are inevitably exposed to X-ray radiation during a CT scan,and this radiation can have a potentially negative impact on a person’s health and safety.Depending on the individual,small doses of X-ray radiation can lead to metabolic disorders,certain genetic diseases and,in more severe cases,cancer,and the effects are more pronounced in children.This is why more and more patients and doctors are becoming concerned about radiation in CT.This can be mitigated to some extent by reducing the dose of radiation during the scan,depending on the patient’s condition.A lower radiation dose may reduce the quality and clarity of CT images,and small lesions may not be clearly visible,increasing the rate of misdiagnosis and affecting diagnostic accuracy.Therefore,it is of great clinical research value and practical significance to study CT(LDCT)image enhancement algorithms under low dose conditions.In this paper,we study the application of image post-processing methods to low-dose CT image enhancement based on real clinical data.The aim is to suppress complex noise in low-dose CT images,enhance the contrast and brightness of the images,and highlight the location of lesions,thus improving the quality of low-dose CT images and making the CT images clearer,which helps doctors in diagnosis and treatment.According to the above problem background,this paper proposes a low-dose CT image enhancement algorithm under generative adversarial network for the research topic of low-dose CT image enhancement.The algorithm consists of three parts,namely lowdose CT image denoising,CT image enhancement and CT image fusion.Firstly,in order to effectively suppress the complex noise in the low-dose CT images and retain the detailed information in the CT images,this paper redesigns the generator,discriminator and loss function based on the generative adversarial network framework of WGAN-GP;secondly,in order to highlight the lesion regions that are not easily observed in the CT images,the multi-scale Retinex with colour recovery and unsharp mask algorithms are used to enhance the brightness and contrast of the denoised CT Finally,in order to fuse the two enhancements more carefully and reduce information redundancy,the enhanced CT image is decomposed into multi-directional subgraphs using a non-downsampling contour transform,and the matched pairs are adaptively fused using a convolutional neural network and inverse transformed to obtain the final enhancement result.This paper uses the official dataset of the AAPM competition as experimental data for denoising,enhancement and fusion,and evaluates the effectiveness of this paper’s method using several objective metrics.The method in this paper achieved results of33.0155,0.9185 and 5.99 in the three common metrics of PSNR,SSIM and RMSE,which exceeded other common algorithms.According to the experimental results,the method in this paper suppresses LDCT image noise while retaining a large amount of detailed information,improves image brightness and contrast,and highlights lesion areas,which helps doctors to accurately analyse the condition.