Denoising Of Digital Breast Tomosynthesis Based On Total Variation In NSCT Domain

Digital Breast Tomosynthesis(DBT)enables three dimensional reconstruction to eliminate interference from overlapping breast tissues.However,DBT images have low contrast and contain noise,which degrades the quality of the image.Therefore,how to effectively denoise the DBT image while preserving the edge contour of the tissue and lesion structure in the DBT image plays a key role in improving the quality of the DBT image.This thesis concentrates on the total variation denoising model and Nonsubsampled Contourlet(NSCT)transform.An improved total variation denoising method and its combination method with the NSCT transform are proposed,which are applied to the denoising of the DBT images.The major work includes the following contents:Firstly,an improved denoising model of Anisotropic Total Variation Weighted Regularization(ATVWR)is proposed.According to the characteristics of the flat and texture regions in the image,the model adaptively adjusts the weights of the horizontal and vertical directions of the regularizations,and then adjusts the filtering strength.The improved model can suppress the "staircase effect" to a certain extent while effectively removing the noise in the image,and can better preserve the texture structures and edge contours in the image.The parameters of the phantom simulation method were improved,and the phantom which can better represent the characteristics of the breast tissue was simulated.Denoising experiments on DBT phantom and practical DBT images show that the proposed model can improve the PSNR value by 1.8dB compared to the existing total variation,related denoising algorithms,and the lesion structure and texture information in the DBT image can be well maintained.Secondly,a denoising algorithm based on NSCT transform and ATVWR model is proposed.The algorithm takes the NSCT coefficients in different directions and different scales of the image as the total variation regularizations,and the total variation weighted regularization operation is performed to minimize the NSCT coefficients,which can effectively optimize the information of different frequency subbands;In addition,the NSCT coefficients can separate the effective information and the noise component,and thus eliminating the "staircase effect" generated by the regularization operation in the iterative process,which can further preserve the edges of the breast tissue,and thus improve image quality.The experimental results show that the proposed method can effectively denoise the DBT images,and retain the microcalcifications and the edges of the tumor.The proposed denoising method can improve the PSNR value by 2.4dB compared with the existing total variation related denoising algorithms.