Region-based SIRT algorithm for the reconstruction of phase bins in dynamic micro-CT

Introduction In dynamic micro-CT, the goal is to visualize the interior of a time-varying object at different points in time. If the scanned object is reconstructed without taking into account the object’s motion, the reconstructed image will be blurred. A standard approach to compensate for this motion is to incorporate a motion model in the reconstruction algorithm. The motion model can be estimated using an external signal, e.g. a signal acquired by the tracking of markers attached to the scanned object, or during the reconstruction, e.g. by optical flow estimates on the intermediate reconstructed images. Another widely used technique is phase binning. In this approach, the motion is assumed to be periodic and all acquired projection data is ordered in bins according to an externally acquired periodic signal that represents the periodicity in the motion. A reconstruction can then be calculated per phase bin. The latter has the disadvantage that its quality depends on the number of acquired projections per phase bin. When scanning small animals, the radiation exposure to the animal should be kept as low as possible. These two facts result in the inevitable trade-off between the animal’s total radiation exposure time and reconstruction quality. However, often there are large regions within the scanned object that remain stationary throughout all phase bins. This extra knowledge can incorporated in the reconstruction algorithm, resulting in comparable image quality with a reduced number of projections. In this contribution, we propose the rSIRT algorithm (region-based Simultaneous Iterative Reconstruction Technique), an algebraic reconstruction algorithm that improves reconstruction quality for phase bin methods. The rSIRT method is demonstrated on a cardiac CT dataset of a mouse.