Trajectory optimization for few-view robot-based CT: Transitioning from static to object-specific acquisition geometries

The advent of robot-based computed tomography systems accelerated the development of trajectory optimization methodologies, with the objective of achieving superior image quality compared to standard trajectories while maintaining the same or even fewer number of required projections. The application of standard trajectories is not only inefficient due to the lack of integration of available prior knowledge about the object under investigation but also suboptimal because of limited accessibility issues during scans of large components, which are common in robot-based computed tomography. In this work, we introduce an object-specific trajectory optimization technique for few-view applications, based on a 3D Radon space analysis using a RANSAC algorithm. In contrast to existing methods, this approach allows for object geometry specific projection views, which are no longer constrained by discretized initial view sets on predefined acquisition geometries. In addition to eliminating the effects of discretized initial sets, this technique offers a distinct advantage in scenarios of limited accessibility by enabling the avoidance of collision elements, unlike trajectory optimizations on predefined acquisition geometries and standard trajectories. Our results show that the presented technology outperforms standard trajectories of evenly distributed projection views on predefined geometries in both ideal accessibility and limited accessibility scenarios. According to the employed geometry-based image quality metrics, our approach allows for reductions of more than 50% in the number of projection views while maintaining equivalent image quality. Robot-based CT enables quality assurance of specific subregions in large components Geometry-relevant projections are essential for assembly analysis Using a 3D Radon space analysis to determine optimal geometry-relevant projections Enhances geometric imaging with fewer projections than standard trajectories