: We have investigated a video-based patient repositioning technique designed to use skin features for radiotherapy repositioning. We investigated the feasibility of the clinical application of this system by quantitative evaluation of performance characteristics of the methodology.
: Multiple regions of interest (ROI) were specified in the field of view of video cameras. We used a normalized correlation pattern-matching algorithm to compute the translations of each ROI pattern in a target image. These translations were compared against trial translations using a quadratic cost function for an optimization process in which the patient rotation and translational parameters were calculated.
: A hierarchical search technique achieved high-speed (compute correlation for 128 × 128 ROI in 512 × 512 target image within 0.005 s) and subpixel spatial accuracy (as high as 0.2 pixel). By treating the observed translations as movements of points on the surfaces of a hypothetical cube, we were able to estimate accurately the actual translations and rotations of the test phantoms used in our experiments to less than 1 mm and 0.2° with a standard deviation of 0.3 mm and 0.5° respectively. For human volunteer cases, we estimated the translations and rotations to have an accuracy of 2 mm and 1.2°.
: A personal computer-based video system is suitable for routine patient setup of fractionated conformal radiotherapy. It is expected to achieve high-precision repositioning of the skin surface with high efficiency.
