ESTRO 2025 - Abstract Book

S3040

Physics - Image acquisition and processing

ESTRO 2025

clinical workflow, possibly due to improved OAR delineations. The automated delineation can be further improved by manual modification, similar to the current CT workflow.

Keywords: MR-only RT, Head-and-Neck cancer. Autodelineations

References: [1] Kaushik, S. S., Bylund, M., Cozzini, C., Shanbhag, D., et al (2023). Region of interest focused MRI to synthetic CT translation using regression and segmentation multi-task network. Physics in Medicine & Biology , 68 (19), 195003. [2] Czipczer, V., Kolozsvári, B., Deák-Karancsi, B., Capala, M. E., et al (2023). Comprehensive deep learning-based framework for automatic organs-at-risk segmentation in head-and-neck and pelvis for MR-guided radiation therapy planning. Frontiers in Physics , 11 , 1236792. [3] Paczona, V. R., Capala, M. E., Deák-Karancsi, B., Borzási, E., et al (2023). Magnetic Resonance Imaging–Based Delineation of Organs at Risk in the Head and Neck Region. Advances in Radiation Oncology , 8 (2), 101042.

3341

Digital Poster The bone rigidity error: a simple, quantitative, and interpretable metric for patient-specific validation of deformable image registration Andreas Smolders 1,2 , Tony Lomax 1,2 , Francesca Albertini 1 1 Center for Proton Therapy, Paul Scherrer Institute, Villigen, Switzerland. 2 Department of Physics, ETH Zurich, Zurich, Switzerland Purpose/Objective: Despite its potential, deformable image registration (DIR) remains underutilized in clinical settings due to challenges in assessing solution quality, particularly in time-sensitive cases. Existing quality metrics can be operator dependent, using manually annotated landmarks or contours, but these are resource-intensive to evaluate and therefore impractical for patient-specific validation. Other metrics are fully automatic, like the Jacobian determinant, measuring local compression or expansion, or consistency metrics. The main limitation of these is that they are challenging to interpret and may be fulfilled even if the deformation is incorrect. Evaluating multiple complementary automatic metrics reduces the likelihood of an inaccurate DIR solution passing all tests. In this work, we propose an additional metric of physical plausibility: the bone rigidity error (BRE). The BRE is based on the principle that, while the transformation is deformable, each individual bone should transform rigidly, even if different bones undergo distinct rigid transformations. Material/Methods: We evaluated the BRE for 6 DIR algorithms on 32 patients with 137 CT-to-CT registrations across 5 datasets spanning all relevant anatomical sites. First, we used the open-source deep-learning tool Totalsegmentator to 1. isolating the vectors of a deformable vector field within each bone; 2. least-squares fitting a rigid registration to these vectors for each bone; 3. calculating the BRE as the average deviation (in millimeters) of these vectors from the fitted rigid registration. A lower BRE indicates better rigidity preservation. Using singular-value-decomposition, this can be implemented in less than 20 lines of code in most modern programming languages. Results: The average BRE varied widely between DIR algorithms, up to a factor 3 for inhale-to-exhale thoracic CT registration (Table 1). Despite large BRE differences between anatomical sites within each algorithm, some algorithms segment each bone individually. The BRE was then calculated by: