SelbyTec | Abstract


Figure: SelbyTec Logo
Selby BP, Walter S, Sakas G, Stilla U (2008) Automatic geometry calibration of X-ray equipment for image guided radiotherapy. In: 47th Conference of the Particle Therapy Co-Operative Group (PTCOG 47): 119.

Automatic Geometry Calibration of X-Ray Equipment for Image Guided Radiotherapy

Purpose: A growing number of commercial health centers employ particle radiation for external tumor treatment. Allowing highly accurate dose delivery onto the carcinogen tissue these techniques are of great benefit for the patients, but also require high precision concerning the patient alignment. Position correction approaches comparing X-ray images of the current patient position to a reference CT series allow final treatment set-up preciseness of less than 1 mm in the tumor area. However, image based patient alignment leads to inaccuracies if the geometric model for the equipment is not defined properly or up-to-date. Especially the accuracy of X-ray based systems with digital flat panels mounted in a rotating gantry suffers from changes of the geometry of the radiographic axes in different gantry positions. Even though the changes of the gantry structure usually are in the sub-millimeter range the effects are visible in the X-ray images acquired for the position correction. If these effects are not modeled in the process of DRR rendering both types of images cannot really be compared, which results in a significant error in the mutual information based automatic position correction approach. To cover all sources of geometric deviation, the calibration has to be performed for several gantry angles and possibly for a set of different snout positions. To enable accurate and reliable patient alignments using image guided procedures, we propose an approach for the automatic calibration of the geometric models of the X-ray equipment and show how the calibrated geometry is used in a patient alignment application.
Methods: The calibration is performed every six months. We use a special designed calibration phantom consisting of a rigid body with a number of equally distributed metallic spheres that can easily be recognized in X-ray images, even from different acquisition angles. The routine itself can be described in several steps. First the phantom is placed on the treatment table with the central sphere located at the isocenter and its axes collinear to the axes of the treatment device. In the next step images are acquired for each beamline at a given set of gantry angles and snout positions. As the device geometry may deviate depending on the rotation direction, each gantry angle has to be approached from two sides. For each X-ray image a model of the metal spheres is computationally projected onto the flat panel plane, using the assumed geometric set-up of X-ray tube and panel. The real sphere positions are segmented from the image by a template matching method. By minimizing the offsets between the assumed and the real positions all nine degrees of freedom (panel shifts and rotations, tube shifts) of the modeled beamline geometry can be adapted to the real geometry. During the alignment correction, the device set-up is given by the treatment plan. Suitable sets of calibration values are chosen from the calibrated configurations and interpolated to minimize the geometric error for the current device setting. The corrected beamline is finally utilized in the computation of DRRs, which are matched to X-ray images to determine the patient alignment.
Results: Tests have been performed in three different gantry constructions using Alderson phantom data for different body parts (head, thorax, pelvis). During the beamline calibration aberrations of the initial geometry could be determined, reaching up to 5 mm panel shift along the main radiographic axis. Without calibrated beamlines the patient alignment procedure was able to achieve accuracies of about ±3.0 mm. With calibrated beamlines the patient alignment accuracy could be increased dramatically so that deviations of less than ±0.5 mm from the perfect alignment could be determined.
Conclusions: Through geometry calibration of the beamlines the accuracy for the patient alignment can be raised dramatically, whereas the initial set-up of the geometry was often not acceptable. Our calibration procedure could reduce patient alignment errors and can additionally serve as an indicator for the geometric accuracy of the treatment device. Comparison of patient alignment results for calibrated and un-calibrated X-ray equipment shows that an adequate calibration routine is indispensable.