Modeling the Anisotropic Resolution and Noise Properties of Digital Breast Tomosynthesis Image Reconstructions.

摘要

In tomosynthesis, a 3D image of the breast is generated from multiple x-ray projections. Although prior authors have modeled image quality in tomosynthesis, their work is limited in approximating the incident angle as normal to the detector. This research studies the consequences of carefully modeling the incident angle. From first principles, an analytical model of the transfer functions for oblique incidence is derived in individual projections. The anisotropy of the transfer functions over the detector area is then demonstrated, and a technique for optimizing detector thickness at various incident angles is developed. In addition to modeling the effect of non-normal incidence on transfer functions, it is shown that oblique incidence yields subpixel translational shifts in the image of the object with each projection angle. By calculating the reconstruction of a high frequency sine input, the potential for super-resolution (i.e., sub-pixel resolution) is established from this property. Superresolution is demonstrated in experimental images of bar patterns and select clinical examples of microcalcifications. The feasibility of super-resolution is also verified in a broad range of multi-planar reconstructions. Finally, from the analytical modeling of the sine input, the scan time which optimizes modulation in the reconstruction is calculated in systems with patient motion.