ESTRO 2025 - Abstract Book

S360

Brachytherapy - Physics

ESTRO 2025

Next, needle reconstruction was performed on the basis of test segmentations for the 19 patients by using existing post-processing methods [2,3] that we extended with additional mechanisms to detect and correct segmentation errors. ANR results were reviewed in Oncentra (Elekta) by an RTT, who modified reconstructions if necessary. A medical physicist subsequently either approved or modified resulting reconstructions further. The time spent by the RTT and medical physicist was measured per patient. We calculated the tip and bottom point deviations comparing ANR, ANR with modifications from RTT and medical physicist (ANRm), and CNR. Results: Results are shown in Figures 1 and 2. Strong agreements were mostly observed. However, substantial divergence also occurred sometimes, beyond the inherent clinical observer variation captured in ANRm vs CNR, especially when the AI model made segmentation errors. Such deviations require expert supervision and adjustment. Processing by the RTT and the medical physicist took on average 12 minutes per patient, including modifications. Most time was spent verifying needles rather than modifying them, with on average <1 minute needed for modifications. Transitioning to ANR-driven reconstruction may therefore require revision of the current clinical workflow to unlock ANR’s time-saving potential.