Session Item

In radiotherapy, PTV margins are used to account for uncertainties. Margin recipes were derived for targets with low curvature which is unrealistic in head and neck (H&N) patients. Robust planning could potentially better spare organs at risk, but the actual robustness to geometrical uncertainties in photon therapy of H&N cancer is unknown. The aim of this study is to compare, for different target shapes, margin-based plans with robustly optimised plans in terms of dose-probability under realistic set-up error scenarios.

24 VMAT photon plans were created on a cylindrical phantom in RayStation with a prescribed dose of 66 Gy, using minimum, maximum and uniform target objectives, and dose fall-off objectives. Margin-based and robustly optimised plans were created for a sphere (48 mm diameter), a cube (48 mm length) and 10 actual CTV shapes from H&N patients (shapes shown in Figure 1), all centred in the phantom. A PTV expansion of 4 mm was used for margin plans and a robustness setting of 4 mm in all cardinal directions for robust plans.

The mean distances between the CTV and the 95% and 50% isodose surface were computed to evaluate conformality. Then, for each plan, 1000 treatments were simulated under set-up uncertainties. Systematic (Σ = 1.2 mm) and random (σ = 1.4 mm) uncertainties were modelled as Gaussian distributions. For each simulation, the minimum cumulative dose to the CTV shape and the D99 was recorded. 

The 90^th percentile of the minimum CTV dose in the simulation results is close to or above the 95% prescribed dose for margin plans of all targets, in accordance with the van Herk margin recipe, but is below this level for all robust plans except for the sphere (Figure 1a). For all plans, the 90^th percentile of the D99 is above the 95% prescribed dose (Figure 1b). This shows that margin plans provide robust coverage for the whole CTV for all shapes yet robust plans fail to cover the entire target in most scenarios. However, regions of high and intermediate dose for the margin plans extend beyond those for the robust plans (Figure 2).  The distance between the CTV and the 95% isodose surface is ~5.9 mm in margin plans, and ~4.1 mm in robust plans.

For all shapes, margin-based plans provided adequate coverage, but high and intermediate dose regions extended further than in robustly optimised plans. The robust plans under-dose small parts of most shapes except for the sphere. This is likely due to the ‘corners’ of the shape not being covered in the high dose region because the robustness algorithm only samples errors in cardinal directions. We conclude that margin plans are safe for all tested shapes, but that robust planning should be used with care because of under-dosage of high-curvature regions in complex CTVs. Better error sampling methods are therefore required.