ESTRO 2025 - Abstract Book

S2896

Physics - Dose prediction, optimisation and applications of photon and electron planning

ESTRO 2025

3749

Digital Poster Voxel-wise estimation of the dosimetric uncertainty of deformable image registration for reirradiation Wolfgang Lechner 1,2 , Maude Cornu 1 , Dietmar Georg 1,2 1 Department of Radiation Oncology, Medical University of Vienna, Vienna, Austria. 2 Christian Doppler Laboratory for Image and Knowledge Driven Precision Radiation Oncology, Department of Radiation Oncology, Medical University of Vienna, Vienna, Austria Purpose/Objective: The number of patients undergoing reirradiation cases is steadily increasing. To accurately estimate the dose distribution from a previous treatment on the current patient’s anatomy, deformable image registration (DIR) algorithms are the method of choice. Despite their continuous improvement, DIR algorithms can introduce significant geometric uncertainties, which translate into respective dosimetric uncertainties. The latter are challenging to quantify but crucial for treatment plan optimization in a reirradiation setting. This study proposes a method for voxel wise estimation of the dosimetric uncertainty resulting from DIR, which not widely available. Material/Methods: Ten patients who underwent re-irradiation in the pelvic region were selected for this study. CT images from both the initial and current treatment courses were registered with two DIR algorithms, i.e. the ANACONDA algorithm as integrated in the treatment planning system (TPS) RayStation (v12B, RaySearch Laboratories AB, Stockholm, Sweden), and the Elastix algorithm implemented in 3D Slicer. The ANACONDA algorithm was applied in three configurations: (1) without controlling regions of interest (cROIs), (2) using whole organs as cROIs, and (3) utilizing the walls of filling organs as cROIs. Additionally, a rigid registration was performed with the TPS. The dose from the initial treatment was mapped to the current CT scan using these five different transformations. The voxel-wise mean dose and standard deviation were calculated using a Python script. The results were converted into DICOM-RT files for subsequent import into the TPS (Figure 1). The highest 1% standard deviation of dose (u D,1% ) was recorded for each organ-at-risk (OAR) and the new PTVs. These differences were tested for significance using the t-test assuming a significance level of 0.05.

Results: Substantial dosimetric uncertainties were observed primarily in regions with steep dose gradients and areas where DIR algorithms exhibited discrepancies (Figure 2). On average, the u D,1% for OARs was 12.7% (range: 6.2% to 21.3%), while it was 9.4% (range: 0.6% to 22.9%) for the planning target volume (PTV) of the current treatment. With exception