A preliminary study of image reconstruction from low-dose data in dedicated breast CT

Abstract

Interest exists in developing tomosynthesis machines and dedicated CT scanners for providing 3D images of breasts. Prototypes of dedicated breast CT scanners have been built and are under evaluation. In these scanners, analytic algorithms such as FDK are currently used for image reconstruction, which generally require data collected at a large number of views over a circular-scanning configuration. Imaging dose to patients is an important concern in breast CT. The current breast CT is designed to deliver about the same total imaging dose as that of a typical two-view mammography exam. This limited total exposure is distributed over a large number of views, thus resulting low-SNR data and noisy images. As breast-tissue contrast is relatively low, high noise level in images can render the tissue-contrast-based diagnosis difficult. Results from both academia and industry in developing iterative algorithms for image reconstruction from diagnostic CT data seem to indicate that they may yield images of higher quality than FDK from low-SNR data. In this work, we have investigated optimization-based iterative algorithms for image reconstruction from low-SNR data in breast CT. We have acquired both physical phantom and patient data for evaluating the utility of breast CT scanner. We formulated the reconstruction problem as a constrained minimization of the image total variation (TV), and used algorithms based upon the ASD-POCS scheme to solve the optimization problem. Using the algorithms developed, we reconstructed images from the phantom and patient data. Because the FDK algorithm is currently used clinically for image reconstruction in breast CT, images reconstructed with the proposed algorithms are compared with those obtained with the FDK algorithm. Results of these studies suggest that optimization-based algorithms may potentially improve image quality over the FDK algorithm for low-SNR breast-CT data.