Beyond 2D Evaluation: MRI Polymer Gel–Based 3D DVH Verification for VMAT Lattice Radiotherapy
Abstract
Purpose
Lattice radiotherapy (LRT) produces steep vertex–valley gradients, making 2D gamma-based QA insufficient for clinically relevant 3D dose–volume endpoints. We propose an MRI polymer gel–based 3D DVH verification workflow for VMAT-LRT and assess feasibility and cross-platform reproducibility using peak-to-valley dose ratio (PVDR) and vertex-level DVH metrics.
Methods
A rigid 12×12×12 cm³ phantom was filled with VIPET-based polymer gel (vGEL). Twenty-seven spherical ROIs (10-mm diameter) were arranged in a 3×3×3 lattice (30-mm spacing). A single-fraction VMAT-LRT plan (12 Gy) was delivered on two linac platforms (TrueBeam and OXRAY) using two independent gel batches (one delivery per machine). Post-irradiation MRI was converted to a 3D dose using in-batch calibration. TPS and gel doses were converted to 3D-DVHs for all spheres and a composite target, and compared using PVDR (D10/D90) and vertex endpoints (D95/D5). Coronal-plane gamma was computed; Gafchromic film was acquired for TrueBeam only.
Results
The gel-reconstructed dose preserved lattice geometry and DVH shape on both platforms. PVDR was 2.65 (plan) and 3.06 (gel) for TrueBeam, and 2.80 (plan) and 3.25 (gel) for OXRAY. Across evaluable vertices (TrueBeam: 23/27; OXRAY: 25/27; exclusions for signal loss), gel doses were higher than TPS: TrueBeam mean ΔD95=+1.0 Gy and mean ΔD5=+1.7 Gy; OXRAY median ΔD95=+1.08 Gy and median ΔD5=+1.94 Gy. Coronal-plane gamma pass rate was 93.4% for vGEL, while film yielded 99.7% (2D, registration-dependent). With gel uncertainty ~3–5% and film ~1–3%, the combined uncertainty (~3–6%; ≈0.4–0.7 Gy at 12 Gy) suggests the larger D5-than-D95 bias is consistent with gradient/registration effects.
Conclusion
Polymer gel dosimetry enables 3D DVH-based verification of VMAT-LRT using interpretable endpoints (PVDR and vertex D95/D5), with consistently higher gel-reconstructed vertex doses than TPS, consistent with possible TPS underestimation at high-gradient vertices. Ongoing work will quantify registration sensitivity via controlled sub-voxel shifts to support standardized inter-institutional reporting.