Modeling acquisition geometries with improved super-resolution in digital breast tomosynthesis

Abstract

In digital breast tomosynthesis (DBT), a reconstruction is created from multiple x-ray projection images. Our previous work demonstrated that the reconstruction is capable of super-resolution (i.e., subpixel resolution) relative to the detector. In order for super-resolution to yield a reliable improvement in image quality, it should be achievable at all positions in the reconstruction. This paper demonstrates that super-resolution is not achievable at all depths, or at all heights above the breast support. For this purpose, a bar pattern phantom was imaged using a commercial DBT system. A goniometry stand was used to orient the long axis of the parallel bars along an oblique plane relative to the breast support. This setup allowed a single test frequency to be visualized over a continuous range of depths. The orientation of the test frequency was parallel to the direction of x-ray tube motion. An oblique reconstruction in the plane of the bar pattern phantom showed that the existence of super-resolution is depth-dependent. To identify design strategies for optimizing super-resolution, a theoretical model was then developed in which a test frequency higher than the alias frequency of the detector was simulated. Two design modifications that improve super-resolution are identified. In particular, it is shown that reducing the spacing between the x-ray source positions minimizes the number of depths lacking super-resolution. Additionally, introducing detector motion along the direction perpendicular to the breast support allows for more uniform super-resolution throughout the image volume. In conclusion, this work presents strategies for optimizing super-resolution in DBT.