ESTRO 2025 - Abstract Book

S3765

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2025

Keywords: Deep Learning,nasopharyngeal carcinoma

References: [1]. Chen YP, Chan ATC, Le QT, Blanchard P, Sun Y, Ma J. Nasopharyngeal carcinoma. Lancet. 2019;394:64–80. [2]. Li WF, Chen NY, Zhang N, et al. Concurrent chemoradiotherapy with/without induction chemotherapy in locoregionally advanced nasopharyngeal carcinoma: long-term results of phase 3 random-ized controlled trial. Int J Cancer. 2019;145:295–305.

1988

Digital Poster Machine learning decision tree models for multiclass classification prognosis after palliative radiotherapy in patient with advanced cancer. Savino Cilla 1 , Romina Rossi 2 , Ragnhild Habberstad 3,4 , Pal Klepstad 5,6 , Monia Dall’Agata 7 , Stein Kaasa 8 , Vanessa Valenti 9 , Erika Galietta 10,11 , Arina A Zamfir 11 , Federica Medici 10 , Costanza Maria Donati 10,11 , Laura Campanacci 12 , Silvia Cammelli 10,11 , Rebecca Sassi 12 , Alberto Bazzocchi 12 , Mira Huhtala 13 , Martijn Boomsma 14 , Francesca De Felice 15 , Simone Ferdinandus 16,17 , Marco C Maltoni 18 , Alessio Giuseppe Morganti 10,11 1 Medical Physics Unit, Responsible Research Hospital, Gemelli Molise Hospital, Campobasso, Italy. 2 Palliative Care Unit, IRCCS Istituto Romagnolo Studio Tumori “Dino Amadori”, Meldola, Italy. 3 Department of Clinical and Molecular Medicine, Norwegian University of Science and Technology, Trondheim, Norway. 4 Department of Oncology, St. Olavs University Hospital, Trondheim, Norway. 5 Department of Circulation and Medical Imaging, Norwegian University of Science and Technology, Trondheim, Norway. 6 Department of Anaesthesiology and Intensive Care Medicine, St Olavs University Hospital, Trondheim, Norway. 7 Unit of Biostatistics and Clinical Trials, IRCCS Istituto Romagnolo per lo Studio dei Tumori (IRST) "Dino Amadori", Meldola, Italy. 8 Department of Oncology, Oslo University Hospital, Oslo, Norway. 9 Palliative Care Unit, IRCCS Istituto Romagnolo per lo Studio dei Tumori (IRST) "Dino Amadori", Meldola, Italy. 10 Radiation Oncology, Department of Medical and Surgical Sciences (DIMEC), Alma Mater Studiorum University of Bologna, Bologna, Italy. 11 Radiation Oncology, IRCCS Azienda Ospedaliero-Universitaria di Bologna, Bologna, Italy. 12 Diagnostic and Interventional Radiology, IRCCS Istituto Ortopedico Rizzoli, Bologna, Italy. 13 Department of Oncology, Turku University Hospital, Turku, Finland. 14 Department of Radiology, Isala Hospital, Dokter van Heesweg 2, 8025 AB, Zwolle, Netherlands. 15 Radiation Oncology, AOU Policlinico Umberto I, Department of Radiological, Oncological and Pathological Sciences, "Sapienza" University of Rome, Rome, Italy. 16 Department of Radiation Oncology, Cyberknife and Radiotherapy, Faculty of Medicine and University Hospital Cologne, Cologne, Germany. 17 Center for Integrated Oncology (CIO), Faculty of Medicine and University Hospital, University of Cologne, Cologne, Germany. 18 Medical Oncology Unit, Department of Medical and Surgical Sciences (DIMEC), Alma Mater Studiorum University of Bologna, Bologna, Italy Purpose/Objective: This study investigates the use of machine learning (ML) decision tree (DT) models to predict survival outcomes in patients with advanced cancer undergoing palliative radiotherapy (PRT) for bone metastases. The aim is to develop multiclass classifiers to estimate the likelihood of mortality at different time points (3, 24, and 52 weeks) following PRT, leveraging clinical, biological, and laboratory data. Material/Methods: This secondary analysis was conducted on data from the Palliative Radiotherapy and Inflammation Study (PRAIS), involving 573 patients. Candidate covariates included 65 variables (clinical, tumor-related, dosimetric, and laboratory). Decision tree models were developed using Python's Scikit-learn library, with hyperparameter tuning via GridSearchCV and cross-validation. Class imbalance was managed using class weighting. Performance was assessed through metrics such as Area Under the Curve (AUC), precision, and recall, with macro- and micro averages calculated for multiclass classification.