ESTRO 2024 - Abstract Book

S4508

Physics - Machine learning models and clinical applications

ESTRO 2024

1934

Proffered Paper

Machine-learning prediction of treatment response to SBRT in oligometastatic gynecological cancer:

Savino Cilla 1 , Maura Campitelli 2 , Maria Antonietta Gambacorta 2 , Raffaella Michela Rinaldi 2 , Francesco Deodato 3 , Donato Pezzulla 3 , Carmela Romano 1 , Andrei Fodor 4 , Concetta Laliscia 5 , Fabio Trippa 6 , Vitaliana De Sanctis 7 , Edy Ippolito 8 , Martina Ferioli 9 , Francesca Titone 10 , Donatella Russo 11 , Vittoria Balcet 12 , Lisa Vicenzi 13 , Vanessa Di Cataldo 14 , Arcangela Raguso 15 , Alessio G Morganti 16 , Gabriella Ferrandina 17 , Gabriella Macchia 3 1 Responsible Research Hospital, Medical Physics Unit, Campobasso, Italy. 2 IRCCS Fondazione Policlinico Universitario A. Gemelli, Radiation Oncology Unit, Roma, Italy. 3 Responsible Research Hospital, Radiation Oncology Unit, Campobasso, Italy. 4 IRCCS San Raffaele Scientific Institute, Radiation Oncology Unit, Milano, Italy. 5 University of Pisa, Radiation Oncology Unit, Pisa, Italy. 6 S. Maria Hospital, Radiation Oncology Unit, Terni, Italy. 7 S. Andrea Hospital, Radiation Oncology Unit, Roma, Italy. 8 Campus Bio-Medico University, Radiation Oncology Unit, Roma, Italy. 9 S. Orsola Malpighi Hospital, Radiation Oncology Unit, Bologna, Italy. 10 University Hospital of Udine, Radiation Oncology Unit, Udine, Italy. 11 Vito Fazzi Hospital, Radiation Oncology Unit, Lecce, Italy. 12 Ospedale degli Infermi, Radiation Oncology Unit, Biella, Italy. 13 Azienda Ospedaliera Universitaria Ospedali Riuniti, Radiation Oncology Unit, Ancona, Italy. 14 University of Florence, Radiation Oncology Unit, Firenze, Italy. 15 Fondazione Casa Sollievo della Sofferenza, Radiation Oncology Unit, S. Giovanni Rotondo, Italy. 16 University of Bologna, Radiation Oncology Unit, Bologna, Italy. 17 IRCCS Fondazione Policlinico Universitario A. Gemelli, Gynecologic Oncology Unit, Roma, Italy

Purpose/Objective:

Nowadays, there is still a paucity of accurate predictive models for clinical outcomes of SBRT treatments for gynecological cancer. We aimed to develop and validate different machine-learning (ML) prediction models for the complete response of oligometastatic uterine cancer after SBRT.

Material/Methods:

One hundred fifty-seven patients with 272 lesions from 14 different institutions and treated with SBRT with radical intent were included in this retrospective study. Thirteen datasets including 222 lesions were combined for model training and internal validation purposes, with an 80:20 ratio. The external testing dataset was selected as the fourteenth Institution with 50 lesions. Lesions that achieved complete response (CR) based on RECIST criteria were defined as responders. Prognostic clinical and dosimetric variables were selected using the LASSO algorithm. Six supervised ML models, including logistic regression (LR), classification and regression tree analysis (CART), and support vector machine (SVM) using four different kernels, were trained and tested to predict the complete response of uterine lesions after SBRT. The performance of models was assessed by receiver operating characteristic curves (ROC), area under the curve (AUC), and calibration curves. An explainable approach based on SHapley Additive exPlanations (SHAP) method was deployed to generate individual explanations of the model's decisions.

Results: