ESTRO 2025 - Abstract Book

S2490

Physics - Autosegmentation

ESTRO 2025

The estimated rectal dose was slightly lower than the planned rectal dose on average as shown from the manual estimate in Figure 2.

Conclusion: Although designed for CT, the AI-generated contours on CBCT demonstrated good agreement with manual contours, with discrepancies primarily observed at the superior rectal boundary. The average difference in dose volume metrics between AI and manual contours was 1.6% ± 2.8%, with a maximum discrepancy of 6.5% in a single patient. These results highlight the potential of AI segmentation on CBCT to efficiently estimate rectum dose, though careful review of AI contours is recommended. This approach could support applications such as clinical audit, decision-making for replanning and offline adaptive radiotherapy.

Keywords: AI, autosegmentation, dose assessment