ESTRO 2024 - Abstract Book

S4352

Physics - Intra-fraction motion management and real-time adaptive radiotherapy

ESTRO 2024

ensured better GTV coverage. Additionally, the PS model required an inference time of less than one minute for predicting six OARs at each fraction.

Figure 1: Example image slice showing the comparison between clinical (left) gross tumor volume (GTV, red), planning tumor volume (PTV, orange), left lung (green), and these segmented by the BM (middle), as well as the PS model (right). Clinical GTV and PTV delineations shown for all panels. Table 1: Preliminary results of three example patients’ dose volume histogram parameters for target volumes and the nearest OARs. These parameters were obtained from clinically delivered doses and re-optimized doses using segmentations generated by both the BM and the PS models.

Conclusion:

Our initial evaluation of OAR segmentations indicate that the PS models could outperform the BMs, with accumulated doses similar to the clinical ground truth when using the contours without modification. The efficiency of the PS model shows potential for practical clinical implementation. We have performed PS training for 11 patients and analysis of dose accumulation for the full cohort is ongoing.