Research On CBCT Image Denoising

As a part of image preprocessing, medical image denoising makes an impact on image post-processing such as segmentation, registration, fusion. The medical image denoising can be divided into two categories:spatial domain denoising and transform domain denoising. The spatial domain denoising methods is represented by classic Gaussian filtering, wiener filtering and emerging Non-Local mean filter, while transform domain is represented by Fourier transform denoising and wavelet transform denoising. Because of its real-time, high sensitivity, convenience of Clinical Application, CBCT imaging system is drawing more and more attention. But for the reason of atomic scattering, the CBCT images contain a lot of noise which decrease the soft tissue contrast and blue the image edge. So it increases the difficulty of clinical diagnosis. How to improve the CBCT image denoising method and reduce the impact of noise on the image accuracy has strong research value and practical significance.We introduce the knowledge of medical image noise, and then we focus on the methods of CBCT image denoising. Based on the model of 3DShepp-Logan, we study the WCMS algorithm in the wavelet domain and non-local means algorithm in the spatial domain and propose our improved algorithm. The main work and innovations are shown as follows:(1)For the complex of the noise and model inaccuracy, we propose a new noise estimation model. Through the model we can simulate actual system noise.(2)Based on the full study of wavelet transform modulus maxima denoising and wavelet threshold denoising, we improved the existing WCMS algorithm. Quite apparent is the fact that dyadic wavelet decomposition is exceedingly directional, mend the wiener filters windows. According to the characters of CBCT image, we proposed the noise variance estimation formula. The experimental results show that it can estimate the CBCT image noise variance more accurate.(3) According to the characteristics of CBCT image and the statistical properties of Gaussian noise, we proposed a denoising method based on the statistical characteristics of CBCT image.(4) Sometimes the image contains some pixels which don't have repeat structure. When denoising, the Non-Local means method will smooth these pixels. We proposed a denoising method based on the non-local means and Multi-scale Dyadic Wavelet Transform. The denoising method can remove noise and preserve the image edge effective.