ESTRO 2024 - Abstract Book

S5136

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2024

combining RF with clinical and dosimetric variables increased the predictive power of the model to an AUC-ROC of 0.82 and balanced accuracy of 77%, indicating strong discriminatory performance in distinguishing between patients with moderate/severe acute xerostomia. A similar pattern can be seen with the overall performance of the three models in predicting SS. The model relying on clinical and mean dose to the parotid gland achieved an AUC-ROC of 0.67 and a balanced accuracy of 66%. Including RF exclusively into the model resulted in an AUC-ROC of 0.78 and balanced accuracy of 67%. Incorporating RF and clinical variables into the model, increased the performance to an AUC-ROC of 0.90, and balanced accuracy of nearly 79%. For this model, the mean parotid dose and Morphological_Compactness1 were both statistically significant in predicting SS. However, an increase in parotid mean dose and a decrease in Compactness1 were only moderately associated with the outcome (OR=1.485, 95%CI=1.05-2.1 and OR=0.116, 95%CI=0.02 - 0.683, respectively).

Conclusion:

Incorporating radiomic analysis of the parotid glands improved the prediction of acute xerostomia and SS for patients with HNC. These results now need to be validated on an external cohort.

Keywords: Xerostomia, Sticky Saliva

2885

Digital Poster

Dosiomics: zoom out the role of dose calculation algorithm in head and neck mandibular toxicity

Lorenzo Placidi 1 , Diego Jurado-Bruggeman 2,3 , Matilde Costa 4 , Sofia Cuttone 4 , Carles Munoz-Montplet 2,5

1 Fondazione Policlinico Universitario Agostino Gemelli, IRCCS, Medical Physics Unit, Roma, Italy. 2 Institut Català d'Oncologia, Medical Physics and Radiation Protection Department, Girona, Spain. 3 Institut d’Investigació Biomèdica de Girona Dr. Josep Trueta (IDIBGI, ONCORFIM, Girona, Spain. 4 Tecnologie Avanzate TA S.r.l, Research and Development Unit, Turin, Italy. 5 University of Girona, Department of Medical Sciences, Girona, Spain

Purpose/Objective:

Prediction models in radiotherapy are nowadays based on multi-parameters analysis, including patient, clinical, imaging, therapy, ‘omics and several other type of (big) data. The aim of this study is to investigate the role of dosiomic features in predicting H&N mandibular toxicity considering different dose calculation algorithms and dose reporting quantities.

Material/Methods: