The influence of large x-ray scatter components in projection images remains a problem for digital breast tomosynthesis, especially when anti-scatter grids may not be used because of dose limitation and possible source/detector geometric limitations. Monte-Carlo simulation of scatter fits better in this situation, but the heavy computational cost hinders its clinical application. To simplify scatter estimation, scatter is often assumed to be smooth. However, scatter is not spatial invariant across a projection image, and where and to what degree the smoothness could be claimed and utilized is unclear. In this study, we investigated this question via multi-resolution analysis based on two experiments: one with direct measurements of scatter profiles in the projection images of an anthropomorphic breast phantom; the other with scatter map obtained from Monte-Carlo simulation that used a voxelized breast model as input. We applied 1D and 2D wavelet-based multi-resolution analyses to the scatter profiles and maps. The first experiment indicated that a reduced number of scatter data points that matches the true data can be extracted from densely sampled but noisy scatter profiles: a data reduction rate of 64-128 was achieved at the inner region of the phantom, suggesting that the slowly changing scatter may be obtained at lower sampling distances of 9.0-17.9 mm. Near the edge of the phantom a data reduction rate of 8 was achieved, corresponding to a sampling distance of 2.2 mm. Similar observations were made from the second experiment.

Date Published: 6 March 2013
PDF: 7 pages
Proc. SPIE 8668, Medical Imaging 2013: Physics of Medical Imaging, 86685L (6 March 2013); doi: 10.1117/12.2007315

Published in SPIE Proceedings Vol. 8668:
Medical Imaging 2013: Physics of Medical Imaging
Robert M. Nishikawa; Bruce R. Whiting; Christoph Hoeschen, Editor(s)