ESTRO 2024 - Abstract Book

S3935

Physics - Image acquisition and processing

ESTRO 2024

patient setup errors (SE) ranging from ±2 to ±4 mm deviations, variations in the tissue density to induce calibration curve (CC) errors from ±3 to ±5% for fat and soft tissues and ±7 to ±11% for bone tissue, and weight changes causing anatomical changes (AC) of ±2 to ±12 mm. The Integral Depth Dose (IDD) in pencil beam PR was simulated using a direct ray-tracing method (OpenReggui) with a multilayer ionization chamber (MLIC) as the detector [3]. The PR simulations were performed on the CT dataset using a 26 x 26 cm 2 field, 210-MeV beam energy, 1 mm spot spacing, and 270° gantry angle. A total of 14503 range error maps (REM) between reference and modified CTs were obtained as the optimal offset for the IDDs along the depth axis (Figure 1). These images were systematically divided into three groups: 10,114 for training, 1,378 for validation, and 3,011 for testing. There was no overlap or mixing of patients between these groups. Specifically, the training set comprised images artificially generated from 22 patients, the validation set from 3 patients, and the testing set from 7 patients. The test patients had previously received in vivo PR measurements in a separate study with a field of 4 x 4 cm 2 using the MLIC detector [2].