Longitudinal Trending of MR-Linac Couch Shift Accuracy and Image Registration Consistency
Abstract
Purpose
Couch shift accuracy and image-registration performance are central to MR-guided radiotherapy, yet QA is often interpreted with fixed tolerances alone. This study applied statistical process control (SPC) to derive machine-specific 2σ (warning) and 3σ (action) limits for a daily couch alignment/shift test incorporating TRUFI auto-registration, and to provide a framework for detecting baseline shifts, particularly after service or recalibration events.
Methods
Daily couch QA records (n=498) from a clinical MR-Linac were retrospectively analyzed. The test used known manual shifts (Lat/Vert/Long = −1/−2/−2 cm) followed by TRUFI auto-registration. Registration accuracy was quantified as Δ = Expected − Reported for each axis and as a 3D magnitude error (Δ3D). Couch reproducibility was quantified by the 3D difference between the return position and the initial plan-enabled couch coordinates (Return Δ3D). Individuals–Moving Range (I–MR) control charts were applied. Center lines (CL) were computed as means, with 2σ warning and 3σ action limits calculated as CL ± 1.77·MR̄ and CL ± 2.66·MR̄, respectively (MR̄ = mean moving range).
Results
For auto-registration accuracy, Δ3D was CL 0.91 mm with 2σ limits 0.29–1.53 mm and 3σ limits −0.03–1.85 mm. Axis-specific CLs were small (ΔLat 0.01 mm, ΔVert −0.44 mm, ΔLong −0.04 mm), indicating minimal systematic bias. For couch return-to-start reproducibility, Return Δ3D was CL 0.73 mm with 2σ limits 0.30–1.15 mm and 3σ limits 0.09–1.37 mm. Notably, the 3σ upper limits for both Δ3D metrics were below a 2 mm (0.2 cm) action tolerance.
Conclusion
I–MR SPC provides practical, machine-specific 2σ/3σ warning/action levels for MR-Linac couch shift and TRUFI registration QA, while also enabling baseline-shift surveillance (e.g., sustained CL shifts or increased MR̄) after maintenance, recalibration, or software changes. The same methodology is readily transferable to other radiotherapy platforms to standardize trending, support post-intervention re-baselining, and strengthen longitudinal performance monitoring beyond tolerance-only QA.