ESTRO 2025 - Abstract Book

S4400

RTT - Treatment planning, OAR and target definitions

ESTRO 2025

References: 1.

Hegarty S, Hardcastle N, Korte J, Kron T, Everitt S, Rahim S, et al. Please Place Your Seat in the Full Upright Position: A Technical Framework for Landing Upright Radiation Therapy in the 21st Century. Front Oncol [Internet]. 2022;12. Available from: https://www.frontiersin.org/journals/oncology/articles/10.3389/fonc.2022.821887 2. Boisbouvier S, Underwood T, McNamara J, Probst H. Upright patient positioning for gantry-free breast radiotherapy: feasibility tests using a robotic chair and specialised bras. Front Oncol [Internet]. 2023;13. Available from: https://www.frontiersin.org/journals/oncology/articles/10.3389/fonc.2023.1250678

4156

Proffered Paper Validation of a deep learning-based automatic contouring algorithm for cardiac substructures in contrast Enhanced CT scans: A Multi-Center Study Axel Licha Radiation Oncology, Insitut Jean Godinot, reims, France Purpose/Objective: Accurate and robust delineation of cardiac substructures is essential for determining dose-effect relationships and reducing the risk of cardiac toxicity in radiotherapy for patients with tumors near critical heart regions. This multicenter study, conducted in collaboration with RaySearch Laboratories (RayStation TPS), sought to validate a deep learning-based automatic contouring algorithm (REF: Vaugier et al. CTRO), evaluating its accuracy and clinical feasibility with two independent external cohorts. Material/Methods: The study analyzed 30 contrast-enhanced CT scans from two cancer centers in France. The algorithm contoured 21 cardiac substructures, including the whole heart, atria, ventricles, coronary arteries (Left Main Coronary Artery, proximal and middle-distal Left Anterior Descending, Circumflex, and Right Coronary Arteries), aortic root, mitral valve, interventricular septum, venae cavae, pulmonary veins, arteries, and ascending aorta. Five radiation oncologists from three centers—comprising one resident, three specialists, and the algorithm’s developer—independently reviewed and scored each substructure based on published definitions. Scores ranged from 3 (no correction), 2 (minimal to moderate correction), 1 (major corrections), to 0 (unacceptable/uncontoured). The analysis reported score distributions, mean review times, and standard deviations. Results: The algorithm successfully contoured all 21 cardiac substructures across the 30 contrast-enhanced CT scans. The score distribution indicated that 55.34% (CI 95% [48.44;62.25]) of substructures required no correction, 47.50% [40.99;54.01] needed minimal to moderate correction, and 11.79% [7.84;15.75] required major adjustments. Substructures with consistently high accuracy scores of 3 included the inferior venae cavae (91.11%), the interventricular septum (88.89%) , and mitral valve (91.11%). Review times varied by substructure: the quickest averaged 21 seconds (SD: 10 seconds), while more complex structures averaged 55 seconds (SD: 64 seconds) for review. The coronary artery segments identified as requiring closer attention included Right Coronary Artery, middle distal segment, Circumflex Artery, middle-distal segment, and Circumflex Artery, proximal segment, with respective mean review times of 55 seconds, 39 seconds, and 36 seconds. The mean correction time for all 21 substructures was 604 seconds (SD: 14 seconds) per patient. These results highlight the algorithm’s clinical potential, with most substructures achieving favorable accuracy and requiring minimal adjustments.(figure 1)