Session Item

On-line adaptive proton therapy is not clinically available. In part this is related to challenges in fast on-line plan (re-)optimization and QA. Recently, we have proposed an on-line adaptive approach for IMPT, based on patient-specific pre-treatment established plan libraries [Oud et al. 2022]. In this approach, for each patient, a set of plans with different setup robustness (SR) settings is generated prior to the treatment. Every fraction, the plan best fitting with the current anatomy, as visualized by repeat CT (rCT) is selected for treatment. Plan selection is based on target coverage achieved by the library plans, with dose distribution recomputed on rCT. This approach outperformed current off-line adaptive proton therapy (APT). To enable clinical feasibility in terms of workload, automated target delineation (AutoContours, AC) on rCTs is needed. We investigated the dosimetric impact of using commercially available target propagation tools for daily AC instead of manual contours for daily online plan library based adaptive IMPT.

Fifteen H&N patients treated with SIB IMPT were included in this study. All patients had manually contoured rCTs (ground truth, GT). For the purpose of the study, we also automatically propagated the CTV7000 and CTV5425 planning contours to each rCT, using the Velocity software (Varian, Palo Alto). Different AC propagation methods were investigated: Rigid, Deformable SinglePass, Deformable MultiPass and Extended Deformable MultiPass, and were eventually combined also with automated post-processing as well as smoothing. Comparison between GT contours and each of 19 AC methods was done on the basis of 1) contour metrics, 2) selected plans in plan library based on-line APT, 3) NTCP differences for xerostomia and dysphagia, and 4) achieved target coverage as simulated in 25 treatments with randomized uncertainties.

AC incorporating smoothing led to significantly smaller targets. Analysis of differences in plans selected while using GT or AC (Fig. 1) led to choice of six most promising AC methods, for which the same plans were selected in ~70% cases and plans with greater SR were selected in ~22% cases, assuring sufficient target coverage. Fig.2A indeed shows similar target coverage for these AC based plans selection for majority of simulated treatments, with smallest differences achieved for AC incorporating MultiPass method. Choice of higher SR plans (~22% cases) lead to slightly higher NTCP (Fig.2B). The NTCP loss was considerably smaller compared to the NTCP gain achieved for plan library approach with GT instead of current off-line APT, and therefore considered as clinically insignificant.

With majority (70%) of plans selected as for ground truth contours, and 22% plans selected with greater SR, automatic target delineation proves to be sufficient for plan library on-line APT. From the investigated methods we would recommend using combination of Rigid and MultiPass methods with post-processing.