Three-Dimensional Fiducial Localization from Adjacent Kv Projections: Minimum Gantry Angle Requirements
Abstract
Purpose
To investigate the feasibility of three-dimensional fiducial localization using small-angle kV image pairs from adjacent projections as a time-efficient alternative to conventional orthogonal imaging, and to determine the minimum gantry angle separation.
Methods
A cube phantom containing a 2mm-diameter fiducial was placed on the Breathing Phantom Gating Assembly (Varian) and imaged using CBCT on Varian TrueBeam. Projection images were extracted using TIGER toolbox. The phantom was aligned to isocenter at the minimum-motion-position (on-isocenter). Between gantry angles of 333.7° and 329.7°, the phantom corresponded to the maximum-motion-position (off-isocenter). The fiducial was automatically detected in each projection image, and 3D localization was performed using an in-house-developed projection-based 3D localization method. Ten projection points were randomly selected at the on- and off-isocenter positions. For each selected projection, a second projection separated by 1, 2, 10, 50, 100, and 200 projections was paired to calculate the three-dimensional fiducial location. Localization accuracy was evaluated by comparing the estimated fiducial position with fiducial positions identified on CBCT images. Based on this analysis, the minimum required gantry angle separation was determined.
Results
From the randomly selected projection points at on-isocenter and off-isocenter positions, the mean angular separations between two selected projections, separated by 1, 2, 10, 50, 100, and 200 projections, were 0.407±0.004°, 0.814±0.004°, 4.054±0.006°, 20.275±0.012°, 40.556±0.011°, and 81.101±0.017°, respectively. For on-isocenter cases, the fiducial localization errors corresponding to separations of 1, 2, 10, 50, 100, and 200 projections were sub-millimeter: 0.33±0.29mm, 0.66±0.87mm, 0.41±0.22mm, 0.51±0.00mm, 0.55±0.09mm, and 0.10±0.22mm, respectively. For off-isocenter cases, the corresponding localization errors were 7.04±5.77mm, 4.34±3.35mm, 1.35±0.69mm, 0.98±0.22mm, 0.89±0.22mm, and 0.70±0.07mm with increasing angular separation, respectively.
Conclusion
For on-isocenter positions, localization errors were within 1mm for all projection separations. For off-isocenter positions, localization error decreases with increasing separation angle, and angular separations of less than 4.054±0.006° resulted in localization errors greater than 2mm.