ESTRO 2024 - Abstract Book

S4547

Physics - Machine learning models and clinical applications

ESTRO 2024

The PCC for test patients before and after calibration (section A). Calibration improves the correlation between real error (absolute error map between sCT and CT) and estimated uncertainty. The PCC after a CT translation shift (3mm on x, y, z axes) (section A). The PCC provides a sensitive measure of the correlation existing between real error and estimated uncertainty. The PCC for patient from unseen center (different from the training center) (section A). The predictions for the other centers exhibit a linear correlation, but it is less robust compared to that observed for the training center. The amplitude of the uncertainty for test patient (section B). A significant difference was observed between patients from the training center and those from another center due to poorer sCT generation (the generation quality is specific to a learning center). Figure 2 shows the CBCT, sCT, absolute error map between sCT and CT and sCT uncertainty heatmap for 4 different patients. Two patients were selected from the training center, while an additional two patients are chosen from another center.