Automatic target delineation for online adaptive H&N IMPT with a plan library approach - A feasibility study

Plan library (PL) based, online adaptive intensity modulated proton therapy (IMPT) was recently proposed for head-and-neck (HN) cancer. In this approach, a set of robust treatment plans is generated prior to the fractionated treatment, each with a different setup robustness (SR) setting. Each fraction, one of the plans is automatically selected for treatment, based on target coverage in the daily repeat-CT (rCT). For clinical implementation, fast and accurate automated target delineation is required. We tested feasibility of using clinically available auto-contouring methods.

For 15 HN patients with in total 67 rCTs with manually generated ground truth (GT) CTV₇₀₀₀ and CTV₅₄₂₅ contours, these contours were also generated with 19 auto-contouring approaches available in commercial software. Auto-contouring was compared to GT regarding contour metrics, selected plans, achieved target coverage and predicted Normal Tissue Complication Probabilities (NTCP).

With the five most promising auto-contouring methods, in 66-72 % of rCTs the same plan was selected as for GT contours, while in 19-26 % of other rCTs, plans with a larger SR were selected, assuring uncompromised target coverage. The 90-percentile near-minimum V_95% for CTV₇₀₀₀ was 96.9 %/>97.6 % for selection based on GT/five best performing auto-contouring methods. For 1/15 patients target coverage was reduced. Auto-contour usage in PL-based on-line adaptive IMPT resulted in small NTCP increases, but these were much smaller than the reported NTCP reductions comparing PL-based on-line adaptive with current off-line adaptive IMPT.

Carefully selected commercially available auto-contouring for plan library based on-line adaptive IMPT generally resulted in adequate patient dose.