Numerical Methods for Coupled Reconstruction and Registration in Digital Breast Tomosynthesis

摘要

Digital Breast Tomosynthesis (DBT) provides an insight into the fine detailsof normal fibroglandular tissues and abnormal lesions by reconstructing apseudo-3D image of the breast. In this respect, DBT overcomes a majorlimitation of conventional X-ray mammography by reducing the confoundingeffects caused by the superposition of breast tissue. In a breast cancerscreening or diagnostic context, a radiologist is interested in detectingchange, which might be indicative of malignant disease. To help automate thistask image registration is required to establish spatial correspondence betweentime points. Typically, images, such as MRI or CT, are first reconstructed andthen registered. This approach can be effective if reconstructing using acomplete set of data. However, for ill-posed, limited-angle problems such asDBT, estimating the deformation is complicated by the significant artefactsassociated with the reconstruction, leading to severe inaccuracies in theregistration. This paper presents a mathematical framework, which couples thetwo tasks and jointly estimates both image intensities and the parameters of atransformation. We evaluate our methods using various computational digital phantoms,uncompressed breast MR images, and in-vivo DBT simulations. Firstly, we compareboth iterative and simultaneous methods to the conventional, sequential methodusing an affine transformation model. We show that jointly estimating imageintensities and parametric transformations gives superior results with respectto reconstruction fidelity and registration accuracy. Also, we incorporate anon-rigid B-spline transformation model into our simultaneous method. Theresults demonstrate a visually plausible recovery of the deformation withpreservation of the reconstruction fidelity.