ESTRO 2025 - Abstract Book

S4404

RTT - Treatment planning, OAR and target definitions

ESTRO 2025

Material/Methods: Materials/methods: Fifteen patients treated with CBCT-based oART for bladder cancer (December 2023-October 2024) were included. One oART fraction (from fractions 9-32, where patient-specific margins were applied) was randomly selected per patient. An expert RTT adapter simulated oART in an emulator. Two simulations were performed for each fraction: one with edited AI-generated bladder and rectum contours combined with an unedited AI-generated contour of the bowel bag (S1), and another with unedited AI-generated contours (S2). The dice similarity coefficient (DSC) for the generated bladder (CTV-T) and volume difference for the bowel bag structures were evaluated between simulations and clinically delivered fractions. Doses to CTV-T and bowel bag were compared. The adaptive session duration (from CBCT acceptance to treatment start) was assessed. Results: Results: Median [IQR] DSC for the CTV-T structure was 0.90 [0.86; 0.92] for S1 and 0.83 [0.77; 0.86] for S2. Median [IQR] absolute volume difference for the bowel bag structure was 39.0 cm³ [13.1; 56.9] for S1 and 69.0 cm³ [37.3;105.9] for S2, corresponding to 2.8% [1.1;5.9] and 4.6% [2.4;8.2] (Figure 1). A median variation in the V30Gy to the bowel bag of 7.6% [3.2;9.8] and 9.1% [2.8;12.5] were observed for S1 and S2, respectively. The median adaptive session duration was 9.30 minutes [8.58;9.48] for S1 and 5.07 minutes [4.55; 5.49] for S2 (Figure 2), compared to 13.8 minutes previously reported [1]. Figure 1. Boxplot presenting absolute volume differences for unedited AI-generated bowel bag segmentations for the two simulated oART workflows.

Figure 2. Boxplots presenting the adaptive session duration for the two simulated oART workflows.