Self-calibrating Projection Geometry for Volumetric Image Reconstruction

We propose a novel method for enabling volumetric image reconstruction from unknown projection geometry of tomographic imaging systems, including CT, cone-beam CT (CBCT), and tomosynthesis systems. The invention enables image reconstruction in cases where it was not previously possible (e.g., custom-designed trajectories on robotic C-arms, or systems using uncalibrated geometries), and more broadly offers improved image quality (e.g., improved spatial resolution and reduced streak artifact) and robustness to patient motion (e.g., inherent compensation for rigid motion) in a manner that does not alter the patient setup or imaging workflow. The method provides a means for accurately estimating the complete geometric description of each projection acquired during a scan by simulating various poses of the x-ray source and detector to determine their unique, scan-specific positions relative to the patient, which is often unknown or inexactly known (e.g., a custom-designed trajectory, or scan-to-scan variability in source and detector position). An image registration framework is used that only requires knowledge of the patient (e.g., prior volumetric image such as a diagnostic CT) and avoids additional workflow complexities by operating independently of fiducial markers or extra hardware. The invention was demonstrated for a C-arm CBCT system and shown to substantially improve image quality over conventional geometric calibration methods.

Category(s):