S2090
Physics - Image acquisition and processing
ESTRO 2026
Figure 2a. Patient 4 presented large differences in the neck flexion; MI-based DIRs did not correct this misalignment, whereas CC-based DIRs aligned the necks but introduced vertebral shearing, Figure 2b. We observed weak trends between treatment interval and DIR performance for some structures (i.e., brainstem), but the dataset size was insufficient to test for significance.A total of 207 manual-autosegmentation pairs were analysed across 2-3 CTs per patient, Figure 1b. Manual segmentations were available (in at least one patient) for 15/31 structures. mDTA values were larger than those observed for DIR.
Digital Poster 4333
Geometric evaluation of deformable image registration for paediatric head and neck reirradiation Ellie Glaister 1 , Chelsea Sargeant 1 , Chelmis Muthoni Thiong'o 1 , Alejandro F Frangi 2,3 , Thomas E Merchant 4 , Marianne Aznar 1 , Eliana Vasquez Osorio 1 1 Division of Cancer Sciences, University of Manchester, Manchester, United Kingdom. 2 Centre for Computational Imaging and Modelling in Medicine (CIMIM), University of Manchester, Manchester, United Kingdom. 3 Division of Informatics, Imaging and Data Sciences, University of Manchester, Manchester, United Kingdom. 4 Department of Radiation Oncology, St. Jude Children's Research Hospital, Memphis, USA Purpose/Objective: Reirradiation is increasingly used in paediatric oncology to manage recurrent/secondary tumours following prior radiotherapy1,2. Reirradiation requires precise understanding of prior doses to minimise side effects. Deformable image registration (DIR) enables dose mapping between treatment courses, yet there remains a clear gap in the evaluation of DIR in the paediatric setting3. In this study we have evaluated the geometrical performance of multiple intensity-based DIRs using a commercial treatment planning system. Material/Methods: Radiotherapy planning data from eight patients who underwent reirradiation for ependymoma (7 infratentorial, 1 supratentorial), selected from a larger cohort1, were included in this study. All patients were between 6 and 9 years old at the time of their initial treatment. Eight DIRs were performed varying similarity measures (mutual information (MI) / cross correlation (CC)), resolution levels (3/5), and smoothing sigma (1/2cm) via RayStation’s scripting interface (v11B-R)4. For geometric performance assessment, we generated structures using RayStation’s Head and Neck CT deep-learning model due to the very limited number of clinical contours available. DIR geometric performance was assessed using mean Distance to Agreement (mDTA) for each structure, and we explored the influence of treatment interval. To provide a benchmark with another common source of uncertainty, manual and autosegmentation contours were compared where corresponding pairs were available. Results: DIR geometric performance showed a trend of higher mDTA for more inferior structures, Figure 1a. mDTAs of structures within the head were below 4mm for most cases, except for the parotid glands and base of tongue. Outliers were observed for patient 8 for the mandible and oral cavity (mDTA 2.7-3.8cm), due to an inconsistent field of view between planning scans,
Made with FlippingBook - Share PDF online