Exploratory regression analysis of the geometric targeting accuracy of the Cyberknife system

Abstract

Quality Assurance is an important part of a safe radiotherapy service. The End to End (E2E) test assesses the overall spatial targeting accuracy of the integrated CyberKnife system for multiple beams delivered isocentrically. The targeting accuracy is quantified based on the dose distribution recorded on film. The complexity of the entire process suggests that data-driven approaches can provide robust solutions for the detection and prediction of errors. Regression analysis is a statistical technique for determining the relationship between a single dependent (criterion) variable (in this analysis: total targeting error) and one or more independent (predictor) variables, which are specific error components from the Anterior-Posterior(AP), Left–Right(LR) or Superior–Inferior(SI) directions. Dominance analysis (DA) was used as a method to find whether one independent variable contributes more variance to the regression effect than another independent variable on average across all possible set of independent variables. We analysed the E2E results for different tracking modalities: Stationary modes (6d Skull, Fiducials, Xsight Spine) and Motion Tracking Modes (Synchrony with fiducials, and XsightLung tracking). In the stationary tracking modes the largest contribution to the total targeting error for the trigeminal, 1path_head and 1path_body path sets was from the LR, AP and AP error components, respectively. For the Motion tracking mode with 1path_body, the most dominant component of the total error was from the SI direction.