Reconstruction for four-dimensional dynamic contrast-enhanced dedicated breast computed tomography (4D DCE-BCT)
Aim and outline of the thesis:
This thesis discusses the technical advancements necessary to accomplish highly accurate dynamic perfusion imaging using breast CT. Chapter one provides a General Introduction to the monitoring of breast cancer treat-ment, dynamic CT imaging, dedicated breast CT, image reconstruction in 3D and 4D, as well as image noise reduction techniques. Chapter two explores theoretically whether a more accurate model of linear attenuation coefficients using Midgley’s decomposition into five coefficients is more useful than classical decompositions into just two components. It comes to the conclusion that Midgley's decomposition is more accurate for the low-energy range of breast CT, but not practical for application in reconstruction. Chapter three presents an optimized protocol for dynamic perfusion imaging in breast CT to achieve high numerical accuracy in the estimation of iodine content. This is made possible while using polychromatic reconstruction from sparse data involving just 40projections. The chapter also discusses limitations causing lower spatial resolution than incomparable static scans. Chapter four describes an approach of combining patient motion data that can be obtained from DCE-MRI sequences to create digital breast CT phantoms with motion. The generated phantoms are then used to create projection data for better understanding of motion artifacts in breast CT. Similar artifacts to the ones observed in real clinical data can be achieved with these simulations. Chapter five introduces a non-rigid motion compensation algorithm for single-phase breast CT scans based on transforming the volume using automatic differentiation to match the reprojected volume and measured projections. The correctness of the method was verified in a simulation study and its performance was evaluated on real clinical cases from two institutions. Overall, the algorithm reduces the appearance of motion artifacts. However, artifacts caused by motion with large amplitude cannot always be fully removed. Chapter six explores the possibility of using approximations to speed up a 4D noise filter by making it possible to execute it on a GPU. Old and new versions of the filter are tested on the same simulated data. The GPU version performs up to 23 times faster. The thesis ends with a General Discussion, containing the author’s opinion about the methods presented in each chapter and their impact on dynamic perfusion breast CT imaging. Moreover, this thesis has a dedicated data management plan, this summary, author's CV, and his scientific contributions during his PhD trajectory.