Multi-Institutional Benchmarking of Stereotactic Targeting Accuracy Using Multi-Target Winston–Lutz Analysis
Abstract
Purpose
Multi‑target Winston–Lutz (Multi‑WL) testing supports stereotactic QA for single‑isocenter, multi‑lesion SRS/SBRT. Targeting accuracy was benchmarked across 10 clinical linacs (Varian and Elekta) used by multiple institutions to identify parameters associated with larger field‑target coincidence errors.
Methods
A Multi‑WL plan was delivered using 8 gantry/couch/collimator combinations imaging six tungsten‑carbide spheres (48 fields/linac). Offsets between the individual field centroids and beam centers were measured for each field/target. Targets included one isocenter and five off‑axis positions, enabling evaluation of off‑axis targeting behavior. For each linac and geometry, mean, median, RMS, and the fractions of fields exceeding 1.0 were calculated. Differences between couch‑kicked (couch 90°/270°) and non‑kicked geometries were estimated with nonparametric bootstrap confidence intervals.
Results
Mean/median/RMS/p95/max coincidence were 0.697/0.600/0.862/1.791/2.81 mm with 17.3% of fields exceeding 1.0 mm. Couch‑kicked deliveries had higher mean coincidence than non‑kicked deliveries (0.879 vs 0.637 mm; mean difference 0.242 mm, 95% bootstrap CI 0.119–0.369). Isocenter targeting was slightly better than off‑axis targeting (0.641 vs 0.708 mm; difference 0.068 mm, 95% bootstrap CI −0.058 to 0.189). Inter‑linac variability was substantial: per‑linac mean coincidence ranged from 0.425 to 1.564 mm. The 3D isocenter vectors (|S|) were 1 mm. The highest‑error unit had 64.6% of fields >1.0 mm (52.1% >1.5 mm), while the lowest‑error unit had 0% >1.0 mm.
Conclusion
Multi‑WL enabled practical benchmarking of stereotactic targeting performance and revealed strong dependence on couch‑rotation geometries. Although 3D vectors passed on all but one linac, per‑field and geometry‑specific results revealed clinically relevant variation not captured by a single 3D metric. Reporting both overall distributions and geometry‑specific metrics can support targeted investigations (e.g., couch walkout or gantry sag).