Geometry-Dependent Tracking Accuracy of an Optical Surface Imaging System for Sgrt Using Symmetric Phantoms
Abstract
Purpose
The accuracy of surface-guided radiation therapy (SGRT) depends on reliable optical surface tracking, which varies with surface geometry. This study evaluates a widely‑used commercial SGRT system by measuring its tracking performance on three symmetric phantoms representing common treatment sites.
Methods
Three phantoms were tested: a flat board (simulating abdomen), a spherical phantom (simulating breast), and a cylindrical phantom (simulating limb). Each underwent controlled displacements (±1–100 mm translations, ±0.5–3° rotations) on a 6‑degree‑of‑freedom couch (PerfectPitch™) as the reference. Absolute errors between system measurements and couch‑encoder values were calculated. Statistical analysis included mean absolute error (MAE), median, interquartile range (IQR), Pearson correlation, linear regression, and Wilcoxon signed‑rank tests.
Results
Tracking accuracy was strongly geometry‑dependent. The spherical phantom achieved sub‑millimeter translational accuracy (MAE0.12mm; median0.08mm, IQR0.17mm) with 96% of measurements within the ≤1mm tolerance. The flat board showed larger translational errors (MAE1.38mm; median0.90mm, IQR1.65mm), with only 47% within tolerance and strong correlation between displacement magnitude and error (r=0.84–0.87, p< 0.001). The cylindrical phantom exhibited intermediate translational accuracy (MAE2.90mm; median1.40mm, IQR3.85mm; 61% tolerance compliance) and moderate displacement correlation (r=0.79-0.82). For rotations, the cylindrical phantom showed acceptable PITCH tracking (median0.20°, IQR0.30°) but poor ROLL performance (median1.15°, IQR1.80°), with most ROLL measurements exceeding ≤0.5°. Flat and spherical phantoms demonstrated high rotational (RTN) errors, with RTN errors (median1.25°, IQR1.05°) exceeding the ≤0.5° tolerance in 92% of cases, indicating a fundamental tracking limitation for these geometries.
Conclusion
The tested SGRT system performed well for translational tracking of spherical breast surfaces but showed unacceptable RTN errors for both spherical and flat surfaces. Cylindrical surfaces showed intermediate translational accuracy, acceptable PITCH tracking, and poor ROLL performance. These geometry-specific limitations necessitate tailored commissioning protocols. Clinical implementation should adopt geometry‑specific tolerances, including supplemental verification, especially for ROLL corrections on cylindrical surfaces and RTN corrections on spherical and flat surfaces.