ESTRO 2024 - Abstract Book

S5768

RTT - Education, training, advanced practice and role developments

ESTRO 2024

Results from four french nationwide delineation workshops: the need for a personalized approach

Vincent Bourbonne 1,2 , Yasmine El Houat 3 , Luc Ollivier 4 , Luc Lafitte 5 , Maxime Gobeli 5 , Audrey Larnaudie 6 , Youssef Ghannam 7 1 University Hospital, Radiation Oncology, Brest, France. 2 INSERM, LaTIM UMR 1101, Brest, France. 3 CHUV, Radiation Oncology, Lausanne, Switzerland. 4 ICO Nantes, Radiation Oncology, Saint-Herblain, France. 5 DLine, DLine, Bordeaux, France. 6 CLCC Baclesse, Radiation Oncology, Caen, France. 7 ICO Angers, Radiation Oncology, Angers, France

Purpose/Objective:

Delineation of target volumes and organs at risk (OARs) is essential in radiation oncology. Depending on the primary tumour, each radiation oncologist delineates several volumes of interest (VOIs): Gross Tumour Volume (GTV) and Clinical Target Volume (CTV). Automatic delineation tools using artificial intelligence (AI) are rapidly developing with several commercial and non-commercial solutions being now available. The vast majority of them achieve poor or moderate results for the definition of target volumes. Training remains essential for both validation of automatic segmentations as well as the delineation by itself when AI-tools fail or are non-existent. Since more than five years, the Société Française des jeunes Radiothérapeutes-Oncologues (SFjRO) launched several delineation webinars and workshops. Analysis of these is generally based on DICE scores or Hausdorff distance. These similarity metrics don’t give any insight on the geometrical distribution of the delineations or the inter-participant discrepancies. Despite delivering the analysis of an expert on the case, results of the workshop remain often poorly exploited and not personalized to the performance of the participants. We hereby provide an overview of new visualization tools that could have a significant impact for the analysis of the delineation workshops and a more focused debriefing. In 2023, the SFjRO has launched 4 delineation cases during a radioanatomy national course : anal cancer, lung cancer, head and neck (H&N) cancer and prostate cancer. Each attending could register to two cases respectively: Head and neck/Prostate and Anal Canal/Lung. Depending on the cases, participants were asked to delineate the tumoral GTV (GTV_T), the nodal GTV (GTV_N), the low or high risk CTV (CTV_LR or CTV_HR). Delineation was realized online with a delay maximum of 3 months and correction occurred during national courses. Overlap between the different segmentations was analyzed using the DICE coefficient. Each delineation was thus compared to the respective expert delineation. Furthermore, a probability map was provided for each case and each VOI. All readers delineations were stacked resulting in a probability map (Stack map) in which the pixel value was equal to the percentage of participants having included it. Two maps were extracted from this probability map: a Ref Stack map and a Stack-Ref map. The Ref-Stack map focused on voxels localized within the expert delineation, cropping the Stack map to the expert VOI. The Stack-Ref map focused on voxels localized outside the expert delineation, by subtracting the reference VOI to the Stack map. A satisfaction survey was sent after the course to every attending. Material/Methods:

Results: