Development and Evaluation of Next Generation Tomosynthesis
Abstract
Digital breast tomosynthesis (DBT) is a three-dimensional x-ray imaging technique used for mammography. Current DBT systems are limited in their screening, diagnostic, and prognostic capabilities and are prone to image reconstruction artifacts such as aliasing. Limitations may be attributed in part to the conventional design’s linear acquisition geometry. Additionally, existing metrics used for evaluating spatial resolution of x-ray imaging systems do not account for aliasing, which has limited the ability to compare novel DBT designs. In this thesis, limitations of conventional DBT are identified and a next generation tomosynthesis (NGT) prototype is designed, constructed, and evaluated to achieve improved spatial resolution and artifact reduction for DBT. The NGT design introduces novel x-ray source and detector motions for acquisition geometry. A Fourier spectral distortion metric (FSD) is developed to evaluate high-frequency spatial resolution in the presence of aliasing, identify spatial anisotropies of tomosynthesis reconstructions, and compare acquisition geometries of tomosynthesis. The FSD has been applied to a commercial DBT system to evaluate the quality of multiplanar reconstruction for DBT. Super resolution and spatial anisotropies are identified in conventional DBT. Spatial resolution for MPR is found to correlate with acquisition geometry and spatial frequency orientation. A rapid-testing simulation platform (PhysicsVCT) has been created to test novel acquisition geometries of the NGT prototype in silico. PhysicsVCTs simulate tomosynthesis image acquisition and reconstruction of physics test objects using ray-tracing. Image reconstructions are evaluated using objective metrics. PhysicsVCTs has been found to agree significantly with physical system results, validating the design of the NGT prototype. Novel acquisition geometries of the NGT system are compared with conventional DBT using phantom experiments and a clinical mastectomy specimen. NGT acquisition geometries show improved spatial resolution, reduced cone-beam- and out-of- plane reconstruction artifacts, improved adipose-fibroglandular tissue contrast, and improved reconstructed image quality. In future research, iterations of NGT can improve tomosynthesis further with multimodal prognostic imaging, stationary 2D x-ray sources, and developments in artificial intelligence.
Subject Area
Bioengineering|Medical imaging|Biomedical engineering|Design|Artificial intelligence|Physics
Recommended Citation
Vent, Trevor Lewis, "Development and Evaluation of Next Generation Tomosynthesis" (2021). Dissertations available from ProQuest. AAI28651887.
https://repository.upenn.edu/dissertations/AAI28651887