CBCT Images Denoising Based On Wavelet And Sparse Representation

In recent years, with the rising incidence of malignant tumors and mortality of the patients, the technology of tumor treatment is improved.Among them, because of rapid developments, the status of radiotherapy technology is improved constantly in the clinical treatment. Image-guided Radiotherapy(IGRT) is considered to be the most advanced precise radiotherapy technology. Cone Beam CT(CBCT) imaging system based IGRT can achieve the goal of improving the accuracy of radiotherapy.CBCT has some advantages such as good real-time performance, high sensitivity, and convenient to use, but CBCT images still have some problems such as low contrast of soft tissue, artifacts and serious noise,especially the weak edges of image will dim due to the noise, it not only hinders the acquiring of position information, but also affects the subsequent clinical operations such as registration and segmentation, consequently increases the difficulty of diagnosis. Noise in CBCT image is mainly affected by the X-ray dose, although increasing the dose of radiation can reduce the noise obviously, the absorbed dose to the patient will also improve, which will be unfavorable to health of the patient.Therefore, under the premise of low dose, it is of great clinical significance to study the denoising method of CBCT image and get high quality image.Firstly, this paper expounds the noise of common medical images and the corresponding knowledge of denoising algorithms, then it focuses on the CBCT images, on the basis of the analysis of the imaging principle of CBCT images, the influencing factors of its quality, about the clustering based sparse representation(CSR) denoising method, proposing two methods of CBCT image denoising based on joint wavelet and CSR. The main work and innovation are as follows:(1) Studying the clustering based sparse representation image denoising algorithm, this method uses the non local self-similarity of the image cluster similar patches and learn dictionaries. Using K-means and Principal Component Analysis(PCA) train the desired dictionary, all the sub-dictionaries constitute a structure sparse dictionary. Compared to the traditional over-complete dictionary, structure sparse dictionary reduces the computational complexity, reduces degree freedom of dictionary assessment, and the sub dictionary for the sparse representation of image patches is more targeted. To a large extent, the adaptation of Dictionary depends on the accuracy of clustering,inaccurate clustering will lead to the loss of detail information.(2) About the CSR based denoising algorithm, using wavelet threshold as a pre-processing, and improving the global threshold, because of the reprocessing of wavelet threshold, high frequency signals of the image are extracted, and the noise influence on clustering accuracy are reduced, improving the detail information of image at the same time, needing less number of clustering, thus reducing the computational complexity of the algorithm, the proposed algorithm protects and improves the detail information.(3) In this work, a denoising method of joint WCMS and CSR is proved. In view of the characteristics that the denoising effect of WCMS Method is poor but its edge protection ability is high, and the denoising effect of CSR method is high but its edge protection ability is poor, combining these two algorithms, taking full advantages of them, obtaining a method which has stronger denoising results.Through the experiment of the tested Shepp-Logan image and clinical CBCT images, the results show that methods we proposed can remove noises better, and at the same time it can preserve the structure and details of the image well, and it can enhance the image contrast, which is helpful to the accurate acquiring of position information and clinical diagnosis.