Study On Compressed Sensing Reconstruction And Denoising For Digital Breast Tomosynthesis Images

Digital breast tomosynthesis(DBT)uses tomography to remove overlap between suspicious lesions and breast tissue when mammography is performed on female breasts.The DBT system takes under-sampled format to obtain projection images.The image reconstruction algorithm directly affects the accuracy of the reconstructed image.Therefore,the correct choice of reconstruction algorithm plays a key role in improving the resolution of the reconstructed image,meanwhile,suppressing the noise effectively is also crucial to the improvement of the resolution of reconstructed images.In this paper,we study the imaging principle of DBT,establish the image reconstruction model based on compressed sensing theory and the denoising model of total variation algorithm combining wavelet and non-local filtering.The major contents include:Building a three-dimensional digital breast phantom and proposing a reconstru-ction algorithm based on compressed sensing theory.The phantom fully simulates the different density tissues in the breast,including glandular tissue,adipose tissue,breast masses,micro-calcifications and unavoidable chest wall tissue during DBT scans.The algorithm is based on the theory of compressive sensing.Firstly we establish reconstruction model with the objective function:the total variation(TV)of the image is taken as a regularization item that makes full use of the sparsity of the image in the gradient domain and preserves the structure information of the image and then reconstructs the image;the error between the projection theoretical calculation value and the actual measurement value as the data fidelity,the image information integrity can be reconstructed effectively and the image reconstruction can be completed with high precision;Secondly the alternating direction methods of multipliers(ADMM)is used to iteratively optimize the objective function of the reconstruction model and to obtain a more accurate optimal solution.The method uses compressed sensing theory to reduce the amount of projection data,the sampling time and dose of X-ray radiation to achieve DBT tomographic reconstruction.Proposing a denoising algorithm for total variation model based on wavelet transform and non-local filter.The algorithm uses the gradient information of the image in different scales in the wavelet domain as a regularization of the model,which can suppress the staircase effect caused by the total variation and the Gibbs phenomenon caused by the wavelet transform while effectively removing the noise;the error between the image to be solved and the noisy image are taken as data fidelity to ensure the information integrity of the denoised image;the split Bregman algorithm is introduced to optimize the objective function with faster convergence rate.Meanwhile,non-local filtering enhances the image details.The proposed method can effectively remove the noise and the structural information of the image is more fully maintained.