ESTRO 2025 - Abstract Book

S62

Invited Speaker

ESTRO 2025

4744

Speaker Abstracts Autosegmentation: Implementation and performance in clinical practice Stine Korreman, Emma S Buhl Danish Center for Particle Therapy, Aarhus University Hospital, Aarhus, Denmark

Abstract:

Recent developments in Artificial Intelligence have improved the quality of autosegmentation to a level where real clinical benefit has become achievable. In breast cancer, deep learning based models are being developed for both organs at risk and for targets. Models may be trained on data of varying quality and diversity, and the performance of models depend on the match between their intended scope and the actual clinical application. This may include factors such as patient group characteristics, guidelines used, imaging configurations, patient immobilization practice, as well as other specific clinical practices. As an example, we will dive into the model developed within the Danish Breast Cancer Group for autosegmentation of targets and heart in locoregional breast cancer radiotherapy. The model has been developed in a national collaboration including data from all clinics in Denmark, and based on a dedicated data set, generated as part of a consensus guideline workshop. The model has been tested quantitatively and qualitatively, and a national randomized trial is presently in preparation for use of the model in clinical practice.

We will furthermore cover aspects of implementation in pilot tests, quality assurance issues, as well as comparison with other published models and implementations.

4746

Speaker Abstracts Automated treatment planning: Implementation and performance in clinical practice Livia Marrazzo Department of Experimental and Clinical Biomedical Sciences "M. Serio", University of Florence, Florence, Italy. Medical Physics Unit, Careggi University Hospital, Florence, Italy Abstract: Breast cancer radiotherapy constitutes a significant portion of a radiotherapy department’s workload, making treatment planning efficiency a key priority. Automated treatment planning has emerged as a promising approach to optimize workflow, reduce planning time, and enhance treatment quality. The use of automation also enhances plan standardization, reducing inter-planner variability and increasing reproducibility across different clinical cases. This presentation explores the implementation of automated planning strategies in clinical practice, their performance compared to conventional manual approaches, and their impact on workflow optimization, with reference to existing literature on automated planning applied to breast cancer treatment. One of the main challenges in automating breast radiotherapy planning is the inherent heterogeneity of target volumes and anatomical variability among patients. The target volume in breast radiotherapy can vary significantly depending on disease stage, tumour biology, risk factors, nodal involvement, and surgical history. While whole-breast irradiation is the most common approach, treatment may be limited to the tumour bed (as in partial breast irradiation) or extended to include nodal regions (axillary levels I–IV and/or internal mammary nodes). Postmastectomy radiotherapy further increases complexity, with target volumes encompassing the chest wall, often influenced by the presence of implants or tissue expanders. This variability translates into a broad range of