S409
Clinical - Gynaecological
ESTRO 2026
predominantly from retrospective datasets. The Least Absolute Shrinkage and Selection Operator (LASSO) method was the most commonly used approach for variable selection (8/31; 25.8%). All studies reported internal validation, but only two (6.5%) included external validation. Substantial heterogeneity was noted in reporting model performance metrics. According to the CHARMS framework, 13 studies (41.9%) were classified as having high, 3 (9.7%) unclear and 15 (48.4%) with low risk of bias. Common biases included low statistical power, inadequate performance reporting, and suboptimal predictor selection methods. Applicability concerns were generally minimal across studies. Conclusion: The application of AI- based predictive modeling in RT for gynecological cancers is constrained by methodological variability, limited external validation, and a scarcity of clinically applicable tools.To enhance clinical translation, future interdisciplinary efforts should prioritize rigorous model development, transparent reporting, comprehensive validation, and clinically interpretable tools ideally through expert consensus Keywords: A.I., Radiotherapy, Predictive model
1 Departement de Radiotherapie, Gustave Roussy, Villejuif, Italy. 2 Nuclear Medicine, IRCCS Azienda Ospedaliero-Universitaria di Bologna, Bologna, Italy. 3 Radiation Oncology, IRCCS Azienda Ospedaliero-Universitaria di Bologna, Bologna, Italy. 4 Department of Medical and Surgical Sciences (DIMEC), Alma Mater Studiorum University of Bologna, Bologna, Italy. 5 Radiation Oncology, Azienda Ospedaliero—Universitaria di Ferrara, Ferrara, Italy. 6 Medical Physics Unit, Responsible Research Hospital, Campobasso, Italy. 7 Radiotherapy Unit, Responsible Research Hospital, Campobasso, Italy. 8 UOC di Radioterapia Oncologica, Dipartimento Diagnostica per Immagini, Radioterapia Oncologica ed Ematologia, Fondazione Policlinico Universitario Agostino Gemelli IRCCS, Roma, Italy. 9 Division of Gynecologic Oncology, IRCCS Azienda Ospedaliero- Universitaria di Bologna, Bologna, Italy. 10 Department of Medical Physics, IRCCS Azienda Ospedaliero-Universitaria di Bologna, Bologna, Italy Purpose/Objective: Post-treatment metabolic response on 18F- FDG-PET/CT is prognostic in locally advanced cervical cancer (LACC). Whether early metabolic response also stratifies the pattern of failure (oligometastatic vs non- oligometastatic) and could guide response- adapted surveillance and salvage is unclear. Material/Methods: We retrospectively analyzed 161 consecutive LACC patients treated with concurrent chemoradiotherapy (CRT) and brachytherapy (BT) boost at a single institution (2007–2021), enrolled in the observational ESTHER study (Ethics Committee CE 973/2020/Oss/AOUBo; written consent obtained). PET response was dichotomized as complete response (CR) versus non-CR (partial response, no metabolic change, or progression). Overall survival (OS) was estimated with Kaplan– Meier and modeled with Cox regression adjusting for FIGO stage (I–IIA; IIB; III–IVA), histology (SCC vs non-SCC), maximum tumor
Digital Poster 739
Stratifying oligometastatic recurrence probability and prognostic profile in locally advanced cervical cancer through FDG-PET response assessment. Federica Medici 1 , Paolo Castellucci 2 , Arina A Zamfir 3 , Ludovica Forlani 4 , Gaia Bergamini 4 , Annamaria Sissi Ferrara 4 , Martina Ferioli 5 ,
Claudio Malizia 2 , Savino Cilla 6 , Milly Buwenge 4 , Gabriella Macchia 7 , Luca
Tagliaferri 8 , Anna M Perrone 9 , Pierandrea De Iaco 4,9 , Lidia Strigari 10 , Stefano Fanti 2 , Alessio G Morganti 4,3 , Silvia Cammelli 4,3
Made with FlippingBook - Share PDF online