ESTRO 2025 - Abstract Book

S3757

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2025

Infermi, Biella, Italy. 13 Radiation Oncology Unit, Azienda Ospedaliera Universitaria Ospedali Riuniti, Ancona, Italy. 14 Radiation Oncology Unit, University of Florence, Firenze, Italy. 15 Radiation Oncology Unit, Fondazione Casa Sollievo della Sofferenza, San Giovanni Rotondo, Italy. 16 Gynecologic Oncology Unit, Fondazione Policlinico Universitario A. Gemelli IRCCS, Roma, Italy Purpose/Objective: To identify potential patient clusters based on pre-treatment clinical and dosimetrical variables able to stratify the treatment response and survival of patients with oligometastatic gynaecological cancer receiving stereotactic body radiation therapy (SBRT) using an unsupervised clustering machine learning method. Material/Methods: This multicentric study was based on a cohort of 14 independent datasets of 172 patients receiving curative-intent SBRT for oligometastatic uterine tumours, yielding a total of 272 lesions. In all institutions, the following clinical and dosimetric variables were collected: age, number of lesions per patients, type of lesion (nodal versus parenchymal), lesion burden (number of treated lesions per patients), treatment site of lesion, number of fractions, total dose, biologically effective dose (BED10) and planning target volume (PTV). An unsupervised clustering method based on K-means algorithm was used to identify clusters of lesions. The obtained groups of lesions were compared in terms of local control (LC), disease metastases-free survival (DMFS) and overall survival (OS). Results: The clustering analysis reported an optimal number of cluster equal to three, as provided by the silhouette coefficient methods (Figure 1, upper). The analysis of variance indicated that the variables contributing the most to the separation of the clusters were the PTV, the BED10 and the type of lesion. In particular, the cluster centroids were found at the values of 54.7 Gy and 16.7 cc, 104.7 Gy and 12.0 cc, 26.5 Gy and 86.5 cc for the clusters 1, 2 and 3, respectively (Figure 1, middle). Significant differences were found between the 3 groups of lesions in terms of LC (p=0.002) and DMFS (p=0.040). At 2 years, LC and DMFS were 74.7%, 84.6%, 47.5% and 22.2%, 27.0%, 48.0% for the clusters 1, 2 and 3, respectively. No differences were found for the OS (p=0.215) (Figure 1, bottom).