The role of multiple anatomical scenarios in plan optimization for carbon ion radiotherapy of pancreatic cancer.

Abstract

To quantify benefits of robust optimization on multiple 4DCT acquisitions combined with off-line treatment adaptation for neoadjuvant carbon ion therapy (CIRT) of pancreatic cancer. For 10 previously treated patients, 4DCTs were acquired around -15 (CT_Plan), -5 (RE₁), -1 (RE₂) and +6 (RE₃) days from RT start. Treatment plans were newly optimized to a dose prescription of 38.4 Gy(RBE) (8 fractions) with a constraint of 38 Gy(RBE) to 1% of the gastrointestinal organs at risk volume (D_1%). Three strategies were tested: (A) robust optimization on CT_Plan maximum exhale (0Ex) with 3 mm set-up, 3% range uncertainty, including 30%-inhale; (B) addition of the RE₁-0Ex scenario; (C) plan recalculation at each RE_i and adaptation (RP_i) according to deviation thresholds from clinical goals. The cumulative variation of target coverage and GI-OARs doses was evaluated. Duodenum contours of all 4DCTs of each patient were registered on CT_Plan-0Ex. The capacity of pre-RT acquisitions to predict duodenum position was investigated by computing the intersection of contours at CT_plan, RE₁, or their union, with respect to subsequent 4DCTs and the CTV, coupled with increasing margin. (A) No recalculation exceeded the D_1% constraint. (B) The inclusion of RE₁-0Ex in the optimization problem improved inter-fraction robustness on a patient-specific basis, but was non-significant on average. (C) Half of the plans would be re-optimized to recover target coverage and/or minimize duodenum dose, at least once. A significant difference was observed between pre-RT duodenum contours when intersecting subsequent contours, either with a margin expansion. Anatomical variations highlighted at multiple RE_i proved that a fast and efficient online adaptation is essential to optimize treatment quality of CIRT for pancreatic cancer.