ESTRO 2025 - Abstract Book

S75

Invited Speaker

ESTRO 2025

This dataset will consist of simulation images (both MRI and CT), daily images (CBCT for Track 1 and MRI for Track 2), RT structure sets, and treatment plans from 10 to 20 anonymized patients. The evaluation within the project is organized around four key tasks. The first task, common to both tracks, involves assessing image quality by aligning the generated sCT images with the reference CT images and computing several image indicators—Mean Error, Mean Absolute Error, peak signal-to-noise ratio, and the structural similarity index measure—across body and bony anatomy. The second task focuses on dose accuracy and involves clinical and “QA plans”, specially designed to uncover local inaccuracies in the sCT images and determine their impact on the resulting dose distribution. For Track 2, a third task has been established to evaluate the inter-fraction variability of Hounsfield Units, with a specific focus on the consistency of bone density measurements across daily sCT images. This task is crucial because it determines whether the sCT algorithm can reliably produce stable HU values over multiple treatment fractions. Meanwhile, Track 1 includes a fourth task aimed at verifying the accuracy of patient positioning. This task assesses the spatial fidelity of the sCT images by comparing shift estimates—derived from aligning daily CBCT images with the sCT images—with known shifts obtained from CBCT and CT alignments. Actually are on-going two projects within Track 1, focused on brain and pelvis. As an open and collaborative effort, the MESCAL project operates independently of any specific vendor or industry influence, ensuring an unbiased approach to improving MRI-only workflows in radiation therapy

4767

Speaker Abstracts Full automation of the MRIgRT workflow: State-of-the-art and clinical perspectives Luise A. Künzel Department of Radiation Oncology and Radiotherapy, Charité – Universitätsmedizin Berlin, corporate member of Freie Universität Berlin and Humboldt Universität zu Berlin, Berlin, Germany

Abstract:

Magnetic resonance imaging-guided radiation therapy (MRIgRT) combines high-resolution imaging with radiation therapy, allowing for precise tumor targeting and real-time monitoring during treatment. The MRIgRT process begins with the preparation of the initial treatment plan based on a dedicated planning computed tomography (CT) and/or MRI to define tumor location and surrounding healthy tissues. Following this, online adaptation occurs, where the plan is adjusted to account for any patient movement or anatomical changes between sessions. The imaging step is crucial for obtaining high-quality MRI scans representing the anatomy of the day, which are then used for contouring—the manual or semi-automated delineation of tumor volumes and critical structures. Next, synthetic CT generation from MRI scans is an essential step to enable accurate dose calculation and plan optimization. Plan optimization refines the treatment plan, aiming to deliver the prescribed dose while minimizing damage to healthy tissue. Online quality assurance (QA) ensures that the treatment plan meets predefined safety and efficacy criteria and ensuring consistency across treatment sessions. Finally, the delivery phase involves administering the radiation dose, where precise alignment and monitoring of the tumor’s position are essential. These steps are critical to ensure the accuracy and effectiveness of the treatment. However, despite its potential, much of the workflow still relies on manual intervention, creating opportunities for automation to improve efficiency and precision. Artificial intelligence (AI) has already demonstrated its potential in several steps of the MRIgRT workflow. For example, AI has been successfully applied in the contouring step, where deep learning algorithms assist in segmenting organs at risk (OAR) from MRI scans. Synthetic CT generation using AI-based models has also been proposed, enabling accurate dose calculation. AI-driven plan optimization has been explored, allowing for faster