| As the number one killer, breast cancer is a threat in the global women’s health. The screening and diagnosis of breast cancer has been facing a severe situation.1970s, a Frenchman named Cros first used x-ray mammography for breast disease inspection, and the mammography achieved a good contrast and high resolution. Since then, the x-ray mammography opened a new beginning for the early detection of breast cancer. During the next40years, from the beginning of the Screen-film Mammography, Digital Mammography to the appearance of Full Field Digital Mammography, mammography experienced a series of improvements with the development of computer science and flat panel detector technology. However, conventional screen-film mammography does not facilitate image manipulation and might have suboptimal contrast-resolution characteristics. Digital mammography and Full Field Digital Mammography, although superior in its image manipulation capability, has an inherent limitation because it depicts complex three-dimensional tissue in two dimensions. Because the pathological structures cover each other in projection image, it is hard for doctors to avoid misdiagnosis and false positive phenomena caused by overlapping structure similar to lesions.Therefore, it is particularly important to put forward a new imaging method that eliminates the overlapping organizations interference to maximum extent.The concept of digital tomosynthesis was first proposed in1972by Grant, which is a technique for producing slice images using conventional x-ray systems. Combining with modern digital image processing technology, it reconstructs an arbitrary number of planes retrospectively from a single acquisition sequence. A finite number of projection images are acquired at varying orientations of x-ray tube, patient and detector to reconstruct these arbitrary planes. The development of digital tomosynthesis was a substantial improvement over computed tomography in that it allowed collecting projection data from limited angles, which is more suitable for imaging specific parts of the body, such as joints, mouth and breast.Combined with DTS imaging technology and the continuous development of flat-panel detector technology, the digital breast tomosynthesis came into being. It is based on a digital mammography generator modified for an x-ray tube motion over an arc relative to the pivoting point that is above the detector surface. Initial investigations with digital breast tomosynthesis have been promising and provide the opportunity to overcome the limitation of conventional mammography by acquiring several views of the breast from different angles and reconstructing the image into a3-D volume set. Digital breast tomosynthesis promises to improve the detection and characterization of lesions by separating from overlapping dense fibro glandular tissue and turns out to be advantageous for calcifications and masses.Many important aspects of breast tomosynthesis imaging can affect its clinical performance. Chief among them is the reconstruction algorithm that generates the representation of the three-dimensional breast volume from the acquired projections. Successful reconstruction algorithms have included traditional shift-and-add reconstruction (simple back projection), analytical algorithm based on Fourier transform and iterative reconstruction algorithm. SAA was first used in DBT imaging as its simple principle and fast reconstruction, but it used to cooperate with different filtering functions to suppress overlapping interference. The most commonly known analytical algorithm based on Fourier transform for tomosynthesis reconstruction is filtered back projection, based on the theory of Fourier slice theorem. It has the advantage of fast reconstruction and suitable for clinical application and disadvantage that demands higher data completeness and imaging geometry. The iterative reconstruction algorithms include statistical iterative method and algebraic iterative method. Statistical iterative method takes the statistical properties of projection into account while algebraic iterative method is to solve underdetermined equations, both of which can introduce regularization for restoration to further optimize image quality. Foreign researchers made fruitful exploration about reconstruction technique for digital breast tomosynthesis. Compared with the in-depth study abroad, the research degree of digital breast tomosynthesis in our country only includes some simple introduction, or foreign literature translation stage at present.Therefore, it is of great significance to further explore imaging technology, especially reconstruction algorithm for digital breast tomosynthesis.A set of breast slices is reconstructed from the limited-angle projections of low dose in digital breast tomosynthesis, which not only improves the image quality but also not significantly increase the radiation dose for breast, thus meeting the public’s demands. With the improvement of people’s self-health awareness, the CT reconstruction also faces a great challenge for both high image quality and lower radiation dose. Acquiring sparse angle projections is an effective way to reduce the radiation dose, so how to achieve a good image using those projections becomes the focus for many research teams.In this paper, after fully learning the foreign literature, we first introduced the hardware components of digital beast tomosynthesis, its imaging principle and some basic reconstruction algorithms, such as SAA algorithm, algebra reconstruction technique algorithm and simultaneous algebra reconstruction technique algorithm. And then we implemented a three-dimensional digital breast phantom and simulate its low dose projections from limited angle to prepare the experimental basis for subsequent reconstruction. At the same time, we completed the existing FBP algorithm and its improved algorithm, verifying the effectiveness and feasibility of FBP algorithm. Finally, aiming at the problem of image reconstruction form incomplete and low-dose projection data for digital breast tomosynthesis and high-quality image reconstruction from projection data at sparse angular views for computed tomography, we improved and implemented two new reconstructions for each other. Details are as follows: First, we develop an analytic breast phantom which projection data can be computed analytically. Researchers and manufacturers abroad have presented clinical images produced by their systems, allowing for qualitative image comparison, but there is no domestic approval digital breast tomosynthesis system in our country, real projection data are difficult to obtain, therefore our research is on the simulation study of experimental data. A similar phantom, the well-known Shepp-Logan phantom, has been used in CT reconstruction work as a standard phantom simulating the human head. But the Shepp-Logan phantom does not reflect the structure of the breast which needs to be spread evenly extruding by compression plate in imaging. So we develop an analytic breast phantom that allows for qualitative comparison of reconstruction algorithms of digital breast tomosynthesis. The phantom consists of simple shapes and aims at capturing the main features of the breast. It includes representations of a pectoralis muscle and fibroglandular tissue regions. In the current implementation, mass lesions and microcalcifications are also included in different tissue backgrounds. We generated projection images from the new breast phantom at limited projection views at equally spaced angular intervals, with the x-ray source covering an arc and the detector remaining stationary. The projection image is calculated in a forward projection algorithm using x-rays through the voxel.Second, having an in-depth overview of the principles of tomosynthesis reconstruction algorithms, we focus on filtered backprojection algorithm and its improved algorithm. The three-dimensional version of the Fourier slice theorem states that when a two-dimensional image of an object is acquired at some particular orientation to that object, the2D Fourier transform of that projection image yields a plane through the3D Fourier space of the object. Because only a limited angle is swept out during one tomosynthesis scan with such a system, information from planes inside a double wedge in Fourier space is obtained. We implement a reconstruction algorithm referring a general theory for filtered backprojection applied to digital breast tomosynthesis, in which, in addition to the usual ramp-type and apodization filters, we introduce a slice thickness filter that dampens the impact of the incomplete sampling of the frequency space due to the limited angular range used in tomosynthesis. The slice thickness filter adjust the weight of integrity data and missing data in Frequency-domain therefore suppressing corresponding ringing in the spatial domain created by discontinuity in the longitudinal direction and controlling the impact of out-of-plane artifacts on image quality. Combining the three filter function, we use the simulated projection data for FBP algorithm programming, experimental results show that, FBP algorithm can reconstruct a slice image of the breast, and using improved ramp function can further improve image quality.Third, we propose and achieve a reconstruction method for digital breast tomosynthesis based on selective total p variation regularization. As the missing projection data in digital breast tomosynthesis, the simultaneous algebra reconstruction technique is often modeled as solving an underdetermined system of linear equations. For the resultant underdetermined linear system, if there is a solution, there will be infinitely many solutions. On the other hand, because the low dose used at each angle in digital breast tomosynthesis imaging, the obtained projection views are much noisier and the image noise is amplified as the number of iterations increases. Total p variation regularization may alleviate the ill-posedness problem in tomosynthesis reconstruction but it often smoothes out important edge information and microcalcifications. Aiming at the exiting problem of image reconstruction, we propose a new reconstruction method based on selective total p variation (TpV) regularization. The new algorithm still aims to minimize an objective function that combines a data fidelity term composed by data error between theoretical calculated value and actual measured value of projections and a selective TpV regularization term. The selective TpV regularization classifies voxels into signal category and noise category by using local gradient information around the voxels and applies different regularization to noise voxels and signal voxels by adaptively selecting the smoothing parameter p. In the framework of adaptive steepest descent-projection onto convex sets(ASD-POCS),the algorithm first uses SART for reconstruction to obtain data consistency as well as to meet the non-negativity constraint, and then performs selective TpV regularization to update image, thereby suppressing image noise and protecting the image edges. Two phase alternates, until iteration convergence stop criteria is satisfied. Experiments on simulated breast phantom are performed. The results demonstrate that the selective TpV method suppresses the image noise and preserves the sharpness of edges, especially is superior for subtle microcalcifications.Fouth, we develop a fast iterative reconstruction method based on the selective total variation for parse angular CT. In recent years, the total variation (TV) minimization method proposed by Sidky has been widely used for parse angular CT as its high image quality, and catches many researchers for further improvements on this basic algorithm. Sparse angle CT reconstruction and DBT reconstruction with limited angle belong to the same problem of reconstruction from incomplete projection data. Considering the effectiveness of selective TpV regularization in the improvement of limited angle DBT reconstruction, we propose a fast iterative reconstruction method based on the selective total variation for parse angular CT. The new algorithm uses the selective total variation instead of the TV regularization, in which the regularization scale is associated with the gradient of the image. The proposed selective total variation adjusts the p which is a smooth function of local gradient of images, referred to as the regularity function. So the different geometric features of various types of signals constraints with different types of diffusion in local area, thereby achieving local information locally adjusted. Meanwhile, inspired by the fast iterative shrinkage/thresholding algorithm (FISTA), in which the initial value of the next iteration is determined by a linear combination of the two previous iterative results, a predication step is incorporated for fast computation in the ASD-POCS framework to improve the low convergence speed causing by algebra reconstruction technique.Two experiments on simulated Sheep-Logan phantom are performed. The results demonstrated that the new method not only improved image reconstruction quality and protected the edge of the image characteristics, but also significantly improved the convergence speed of the iterative reconstruction, which has a great significance for clinical application.In this paper, we have achieved some initial results about image reconstruction for digital breast tomosynthesis and sparse angle CT, but the learning is endless, getting some real clinical data to start a more detailed exploration, designing some more effective filtering function for the FBP algorithm, will be our next important work. |
PDF Full Text Request