ESTRO 2025 - Abstract Book

S2529

Physics - Autosegmentation

ESTRO 2025

Material/Methods:

This study used data from 140 stage III NSCLC patients treated with chemo-radiotherapy, including a planning CT (pCT), and a CBCT at each fraction. As shown in Figure 1, the CBCTs are first converted to synthetic CTs (sCT), by applying a previously developed AI model, resulting in superior quality. Thereafter, a 3D UNet model, based on the nnUNet framework [2], was trained to segment the 3D sCT images into muscle groups, adipose tissues, heart, and lungs. A five-case training dataset was manually segmented by experts and used for training. The prediction quality