Purpose: To determine the effects of imager source and panel positioning uncertainties on the accuracy of dual intensity-based 2D∕3D image registration of cranial images.
Methods: An open source 2D∕3D image registration algorithm has been developed for registration of two orthogonal x-rays to a 3D volumetric image. The initialization files of the algorithm allow for nine degrees of freedom system calibration including x, y, z positions of the source and panel, and three rotational degrees of freedom of the panel about each of the three translational axes. A baseline system calibration was established and a baseline 2D∕3D registration between two orthogonal x-rays and the volumetric image was determined. The calibration file was manipulated to insert errors into each of the nine calibration variables of both imager geometries. Rigid six degrees of freedom registrations were iterated for each panel or source positional error over a range of predetermined calibration errors to determine the resulting error in the registration versus the baseline registration due to the manipulated error of the panel or source calibration.
Results: Panel and source translational errors orthogonal to the imager∕panel axis introduced the greatest errors in the registration accuracy (4.0 mm geometric error results in up to 2.7 mm registration error). Panel rotation about the imaging direction also resulted in errors of the registration (2.0° geometric error results in up to 1.7° registration error). Differences in magnification and panel tilt and roll, i.e., source and∕or panel translation along the imaging direction and panel rotations about the orthogonal axes had minimal effects on the registration accuracy (below 0.3 mm and 0.2° registration error).
Conclusions: While five of the nine imaging system variables were found to have a considerable effect on 2D∕3D registration accuracy of cranial images, the other four variables showed minimal effects. Vendors typically provide simplified calibration procedures which aim to remove encountered geometric uncertainties by accounting for two panel translations. This study shows that at least the five relevant positional variables should be separately calibrated, if accurate alignment is required for 2D∕3D registration.