Simultaneous calibration for cone-beam computed tomography

Purpose: Tomographic image quality crucially depends on a precise knowledge of the scan geometry of the acquisition system, especially in micro-CT (μCT). But also C-arm systems require monitoring the focus-detector alignment in order to compensate for imperfections (misalignment) prior to image reconstruction. We developed a method to calibrate rotational cone-beam systems simultaneously with each individual scan without sophisticated calibration phantoms. Methods and Materials: Tracing the projection orbits of point objects, we determine 5 of the 6 parameters describing orientation and placement of the detector relative to the focus plus the point coordinates inside the probe. Our algorithm iterates explicit analytic expressions for the parameters and the coordinates to solve the underlying non-linear optimization problem. It assumes that the angle (tilt) between detector and rotation axis is known (e.g., zero). We simulated rotational (2π) scans of point objects (with noise) on misaligned detectors to compare the reconstructed parameters/coordinates to the known true ones. We measured on the μCT scanners at our institute, and on a commercial μCT system (VAMP, Moehrendorf/Germany), to assess the reduction of image artifacts after misalignment corrections. Vertical arrays of small bearing balls were scanned with spatial resolution ranging from 10 - 100 μm. After segmentation and absorption-weighted center determination we obtained up to 20 complete point orbits on the detector area which were then used in our misalignment analysis. Results: The simulations yielded that we recovered the true skew and slant of the detector, its horizontal and vertical displacements, and the focus-detector distance with errors below 1%, 2%, 1%, 3%, and 1%, respectively. Up to several degrees of detector tilt did not lead to noticeable artifacts and were neglected. Increasing the number of points increased the accuracy and provided a good estimate of residual errors. Since their locations were also reconstructed, the method proved insensitive to phantom specifications and placement uncertainties. In the measured μCT scans image artifacts were greatly reduced. Moreover, the accuracy of the method allowed to assess other corrections, e.g., for detector distortion. Conclusion: Misaligned mechanically stable scanning devices can be calibrated with simple point objects. For systems with unstable geometry, e.g., due to focal-spot motion, markers can be added to the actual probe, or the object holder, such that no extra calibration scan is necessary.http://rsna2003.rsna.org/Rsna2003/Vbk/Conference/event_display.cfm?id=66601&em_id=310084