S585
Clinical – Head & neck
ESTRO 2026
Conclusion: The geometric accuracy of the head and neck’s OARs varies across ABAS software. The vast majority impact of different software is observed in small-volume OARs. However, it proved to be a possible option to reduce the elapsed contouring time for clinical OARs in nasopharyngeal radiotherapy. Keywords: Autosegmentation, Head and neck, Organ at Risk Comparison between planning MRI and contrast- enhanced planning CT for Gross Tumor Volume delineation in primary radiotherapy of Head & Neck carcinoma Adelina Brezae 1,2 , Nikos Paragios 2 , Eric Deutsch 3 , Panagiotis Balermpas 4 , Charlene Bouyer 1 , Ricardo Dal Bello 4 , Peter Kuess 5 , Audrey Maury 6 , Frank Pilleul 1 , Gaber Plavc 7 , Melissa Scricciolo 8 , Vincent Gregoire 1 1 Department of Radiation Oncology, Centre Léon Bérard, Lyon, France. 2 Clinical Affairs, Therapanacea, Paris, France. 3 Department of Radiation Oncology, Institute Gustave Roussy, Villejuif, France. 4 Department of Radiation Oncology, Zurich University Digital Poster Highlight 2068 Hospital, Zurich, Switzerland. 5 Department of Radiation Oncology, Medical University of Vienna, Vienna, Austria. 6 Department of Radiation Oncology, CHU de Saint Etienne, Saint Etienne, France. 7 Department of Radiation Oncology, Institute of Oncology Ljubljana, Ljubljana, Slovenia. 8 Department of Radiation Oncology, Ospedale Dell’Angelo, Venice, Italy Purpose/Objective: Accurate Gross Tumor Volume (GTV) delineation in primary head and neck (H&N) radiotherapy is critical for treatment planning. However, anatomical complexity challenges accurate and reproducible GTV delineation. While contrast-enhanced CT (ceCT) remains standard for radiotherapy planning, MRI offers superior soft-tissue discrimination. This study evaluates MRI's anatomic added value in H&N GTV delineation by comparing T1-, T2-weighted, and T1- gadolinium-enhanced (T1G) sequences to ceCT, assessing tumor volume and intra- and inter-observer variability (OV) among 4 radiation oncologists (RO). Material/Methods: From a total of 190 enrolled H&N carcinoma patients (stage II-IV) from four institutes (Centre Léon Bérard Lyon (CLB), Gustave Roussy Villejuif (GR), University Hospital Zurich (UHZ), Medical University Vienna (MUV)), 90 were so far evaluated. Of those 17 were primarily larynx/hypopharynx, 17 nasopharynx, 46 oropharynx and 10 others.Patients underwent simulation imaging with both ceCT and either 1.5T
contour satisfaction were scored by clinicians. Results: Atlas algorithm in each software affects OARs contouring performance, while increasing numbers of patients in the library did not increase the contour geometric accuracy. The Atlas library of 20 patients showed insignificantly different performance from 40, 60, 80, and 100 Atlas patients. A high geometric accuracy of OARs contour generated from ABAS was found in RayStation (DSC=0.71±0.08, MDA= 0.10±0.04, and HD= 0.57±0.21), followed by MIM and Velocity. RayStation and MIM showed an outperformance over Velocity by achieving DSC higher than 0.7 or MDA lower than 0.3 mm in all structures as per the tolerance limit by AAPM TG132. The HD for large volume structure is insignificantly different among ABAS software, while a larger HD was observed in a small volume structure generated from Velocity, especially in the optic chiasm, optic nerves, and Cochlea. The percentage of clinical satisfaction of OARs contours generated from ABAS was 94%, 86%, and 54% for RayStation, MIM, and Velocity. The elapsed time of OARs auto-segmentation with modification in all software was significantly less than manual contouring (p<0.01). The average time for contour modification was 17.18±3.09, 24.26±5.62, and 23.82±4.67 minutes for RayStation, MIM, and Velocity.
Made with FlippingBook - Share PDF online