Session Item

To investigate the possibility of training a Knowledge-Based (KB) model by a robust auto-planning technique for the right-sided (R) whole breast treatment with ViTAT (Virtual Tangential Arc Therapy).

Since 2020, an automatic KB-planning approach using a volumetric technique ViTAT, that mimics conventional Tangential Field (TF) for R-irradiation, was clinically implemented fully replacing manual TF optimization. ViTAT plans consisted of 4 arcs (6MV) optimized with a VMAT technique (Rapid Arc, Varian) with start/stop angles consistent with TF geometry (60°/220°); arcs were completely blocked, apart from the first and last 20° of rotation. The implemented KB-model was based on 80 clinical TF plans (TF-model) and trained by the RapidPlan (RP) tool implemented in the Eclipse TPS (v13.6, Varian): all patients were treated to 40Gy in 15 fractions. The same patients were re-optimized by the Auto-Planning (AP) module implemented in the Pinnacle TPS (Philips) using intensity-modulated fixed 6MV tangential beams, according to an optimized template developed in another Institute. Beam angles were manually chosen for each patient, while planning was done with no further manual intervention. The resulting dose distributions were used to build a new KB-model using RP (AP-model), aiming to stress the performances of ViTAT. Validation tests were performed on 20 new patients for both KB-plans optimized with AP-model (APm-KBP) and TF-model (TFm-KBP). KB-ViTAT plans were compared in terms of OARs/PTVs dose-volume parameters. Wilcoxon-tests were performed to assess statistically significant differences (p<0.05).

The goodness of the regression model, indicated by the R² parameter, was higher for the AP-model for all OARs. Of note, the R² for AP-model/TF-model was 0.63/0.60 for the ipsilateral lung; 0.69/0.44 for the heart; 0.53/0.44 for the contralateral breast and 0.61/0.55 for the contralateral lung, reflecting the absence of any inter-planner variability in the AP-model. Concerning planning performances, APm-KBP resulted in a better sparing of the ipsilateral lung with respect to TFm-KBP, while coverage and other OARs were similar (figure 1). PTV V95% was 96.5/96.7% for APm-KBP/TF-m-KBP, respectively; D1% was 41.8/41.9 Gy. The ipsilateral lung V20Gy was 12.8/13.7% (p=0.000007), V35Gy was 4.5/5.5% (p=0.000001) and the mean dose (Dm) was 6.5/7.0Gy (p=0.0001). On average, APm-KBP contralateral OARs were equivalent to TFm-KBP.

We demonstrated the possibility of further improving KB-planning for tangential-like whole breast radiotherapy using a model trained with a more consistent and robust technique obtained by auto-plan optimization. The APm-KBP was able to generate automatic plans better than the KB based on clinical ones. Using APm-KBP it was possible to significantly improve the sparing of the ipsilateral lung DVH-tailor, without any compromise of the PTV coverage.

Acknowledge: The study is supported by an AIRC grant (IG23150).