Low Dose CT Image Post-processing Algorithm

Low-dose computed tomography(CT) can reduce the radiation dose to patients.However, with the reduction of radiation dose, the number of photons reaching the detector probe will be lower and then led to poor quality of the reconstructed image. In recent years,researchers has paid widely attention to the research of reconstructing low noise and high resolution CT images in the low-dose scanning conditions. Currently, one of important directions of low-dose CT(LDCT) is the image post-processing algorithms of noise in reconstruction image and the correction of artifacts. The important features of these algorithms are the independent of projection data, which can reduce the noise of LDCT images directly. The post-processing algorithms of LDCT images should preserve the original image details as much as possible, while suppressing noise. This paper mainly focused on LDCT post-processing algorithm, the main work is as follows:Based on the traditional Perona-Malik(PM) model, an improved anisotropic diffusion model of LDCT post-processing algorithms was proposed to overcome the disadvantage of traditional PM model which add the ladder artifacts to the processed images. The new algorithm was combined with intuition fuzzy entropy and the weighted variance. The intuition fuzzy entropy can distinguish between smooth areas and detail areas effectively in images.The weighted variance method can preserve more details while removing image noise.Combining the intuition fuzzy entropy and the weighted variance method, a new anisotropic diffusion coefficient was formed to remove the image strip artifacts while preserving image edges and more details and texture structure.In this paper, a kind of nonlocal average de-noising algorithm with the Gaussian filter is proposed to remove the strip artifacts and noise effectively in the LDCT image. The new algorithm applied the Gaussian filter to preprocess LDCT image at first. Then, the filtered image was used to obtain the new Euclidean distance, which improve the accuracy of thesimilarity measure of the processed images. Experimental results show that, the new algorithm overcomes the shortage of the original nonlocal average de-noising algorithm. The proposed algorithm not only can remove the image artifacts and noise, but also can retain more image edges and details.In this part, based on the total generalized variation, a kind of LDCT image post-processing algorithm was proposed. Firstly, the LDCT images is decomposed by wavelet.Secondly, the anisotropic diffusion method for the different scales in the wavelet domain was applied to process the decomposition coefficients of low-frequency, horizontal, vertical and the diagonal directions of images. Finally, we use the total generalized variation regularization denoising method to supress the wavelet domain reconstructed images strip artifacts.