Combined Reconstruction and Registration of Digital Breast Tomosynthesis

Abstract

Abstract. Digital breast tomosynthesis (DBT) has the potential to enhance breast cancer detection by reducing the confounding effect of superimposed tissue associated with conventional mammography. In addition the increased volumetric information should enable temporal datasets to be more accurately compared, a task that radiologists routinely apply to conventional mammograms to detect the changes associated with malignancy. In this paper we address the problem of comparing DBT data by combining reconstruction of a pair of temporal volumes with their registration. Using a simple test object, and DBT simulations from in vivo breast compressions imaged using MRI, we demonstrate that this combined reconstruction and registration approach produces improvements in both the reconstructed volumes and the estimated transformation parameters when compared to performing the tasks sequentially.

dPIRPLE: a joint estimation framework for deformable registration and penalized-likelihood CT image reconstruction using prior images

“…Second, MBR allows the addition of other sources of information to enforce or encourage reconstructions with desirable image properties. For example, image reconstruction can be regularized using assumptions on image smoothness (Lange, 1990), edge-preservation (Vogel and Oman, 1996; Sidky and Pan, 2008), and self-similarity of the object via dictionary learning (Qiong et al , 2012; Yang et al , 2012). However, additional sources of information used in MBR usually entail only very general properties of the reconstructed image and typically do not incorporate patient-specific prior information.…”

“…The idea of joint estimation has been widely studied in MBR in many modalities. (Gilland et al , 2002; Chun and Fessler, 2009; Fessler, 2010;Yang et al , 2013; Taguchi et al , 2007) For example, in cardiac gated ECT, Gilland et al designed an objective function that jointly estimated cardiac images at two different frames and the cardiac motion between the two frames (Gilland et al , 2002), thereby solving both deformation and attenuation together using a conjugate gradient method. In PET, Fessler used an objective function that jointly estimated a single image (at a specific time point) and a series of deformation fields (used to match motions at other time points), achieving a joint solution by alternately updating the deformation and attenuation parameters.…”

“…(Fessler, 2010) In digital breast tomosynthesis, Yang et al presented a framework that coupled the limited-angle reconstruction of the breast and registration of two temporal breast images and showed that the joint method gave superior results with respect to reconstruction fidelity and registration accuracy than traditional sequential method. (Yang et al , 2013) Despite the varied use of joint estimation in MBR, the joint approach has received less attention in PIBR.…”

Joint Registration and Limited-Angle Reconstruction of Digital Breast Tomosynthesis

“…Rather than perform the two tasks sequentially or iteratively [7,8], we propose a fully coupled algorithm using a simultaneous reconstruction and registration framework summarised in Algorithm 1. The objective function is given by…”

Multi-atlas propagation based left atrium segmentation coupled with super-voxel based pulmonary veins delineation in late gadolinium-enhanced cardiac MRI

“…Comparisons of the manually drawn ground truth (purple boundary) (d), initial automatic segmentation (green boundary) (e), and final segmentation (cyan boundary) (f) are also shown. Table 1 tabulates the Dice score (0.91±0.08) and the Jaccard index (0.84±0.10) 14 of the final segmentation calculated with respect to the ground truth. Figure 3 (a) shows that our final segmentation has significant improvements compared to the initial segmentation based on the WHS alone.…”