We present two new approximate variants of motion compensated 3D back-projection filtration (MC-BPF) algorithms for circular cone-beam X-ray tomography in divergent beam geometry. The first one uses differentiation in a single projection (SD), while the second one is based on the Katsevich-type differentiation (KD) involving two neighboring projections. The BPF type algorithms are compared to an approximate 3D motion compensated filtered back-projection (MC-FDK) algorithm investigated earlier. This work is driven by the hypothesis that shading artifacts observed in the approximate 3D MC-FDK algorithm are mitigated by BPF-type algorithms due to a more local correction of the motion. The MC-BPF consists of three steps: (1) short range differentiation filter, (2) motion-compensated backprojection and (3) inverse Hilbert filtering of the DBP. The long range Hilbert filter is applied to the already motion-corrected DBP. In a simulation study using the Forbild thorax phantom with beating heart insert, a slow and a fast acquisition are investigated. The motion-compensated reconstructions for the fast acquisition show almost the same image quality as the reconstructions from the static reference. The motion-compensated reconstructions for the slow acquisition are capable to recover high and medium contrast objects, however artifacts remain especially at the transition edges towards the lungs. Only minor differences in the quality between the MC-FDK, MC-BPF-KD and MC-BPF-SD are observed.
In 1993 the owner of the Val des Dir dam in Switzerland decided to increase the peak energy production by adding a new power station to the existing storage lake. In the article, the history of the project, the specification for the generators as well as the technical solution provided are described. Further on, the first results of the measurements done during commissioning are presented. The generators in the Bieudron power station are the world's largest fully water cooled machines in operation.
It is well known that rotational C-arm systems are capable of providing 3D tomographic X-ray images with much higher spatial resolution than conventional CT systems. Using flat X-ray detectors, the pixel size of the detector typically is in the range of the size of the test objects. Therefore, the finite extent of the "point" source cannot be neglected for the determination of the MTF. A practical algorithm has been developed that includes bias estimation and subtraction, averaging in the spatial domain, and correction for the frequency content of the imaged bead or wire. Using this algorithm, the wire and the bead method are analyzed for flat detector based 3D X-ray systems with the use of standard CT performance phantoms. Results on both experimental and simulated data are presented. It is found that the approximation of applying the analysis of the wire method to a bead measurement is justified within 3% accuracy up to the first zero of the MTF.
Cardiac C-Arm computed tomography leads to a view-starved reconstruction problem because of electrocardiogram gating. Reconstruction with the Feldkamp, Davis and Kress method (FDK) generates large streak artifacts in the reconstructed volumes, hampering the medical interpretation. In order to remove these artifacts, deconvolution techniques have been proposed. In this paper, we start from a recent inverse filtering method developed for streak artifact removal. Two ways to improve upon this method are described. It is then proposed to replace inverse filtering with an iterative deconvolution scheme. Finally, we show that the iterative deconvolution method can itself be replaced by iterative filtered back projection (IFBP). The IFBP approach is flexible and could be used in a broad range of applications, while the improved inverse filtering approaches are computationally less demanding and better suited for time-critical applications.
A fully automated 3D centerline modeling algorithm for coronary arteries is presented. It utilizes a subset of standard rotational X-Ray angiography projections that correspond to a single cardiac phase. The projection selection is based on a simultaneously recorded electrocardiogram (ECG). The algorithm utilizes a region growing approach, which selects voxels in 3D space that most probably belong to the vascular structure. The local growing speed is controlled by a 3D response computation algorithm. This algorithm calculates a measure for the probability of a point in 3D to belong to a vessel or not. Centerlines of all detected vessels are extracted from the 3D representation built during the region growing and linked in a hierarchical manner. The centerlines representing the most significant vessels are selected by a geometry-based weighting criterion. The theoretically achievable accuracy of the algorithm is evaluated on simulated projections of a virtual heart phantom. It is capable of extracting coronary centerlines with an accuracy that is mainly limited by projection and volume quantization (0.25 mm). The algorithm needs at least 3 projections for modeling, while in the phantom study, 5 projections are sufficient to achieve the best possible accuracy. It is shown that the algorithm is reasonably insensitive to residual motion, which means that it is able to cope with inconsistencies within the projection data set caused by finite gating accuracy, respiration or irregular heart beats. Its practical feasibility is demonstrated on clinical cases showing automatically generated models of left and right coronary arteries (LCA/RCA).
Label ranking is a specific type of preference learning problem, namely the problem of learning a model that maps instances to rankings over a finite set of predefined alternatives. These alternatives are identified by their name or label while not being characterized in terms of any properties or features that could be potentially useful for learning. In this paper, we consider a generalization of the label ranking problem that we call dyad ranking. In dyad ranking, not only the instances but also the alternatives are represented in terms of attributes. For learning in the setting of dyad ranking, we propose an extension of an existing label ranking method based on the Plackett-Luce model, a statistical model for rank data. Moreover, we present first experimental results confirming the usefulness of the additional information provided by the feature description of alternatives.
