ESTRO 2021 Abstract Book

S510

ESTRO 2021

compared to TNM 8 th edition, the gold standard. Materials and Methods

Data of 666 retrospective (training) and 143 prospective (validation) locoregionally advanced (stage III‐IVA/B (M0) according to TNM7) HNSCC patients were collected. A multivariable Cox regression model was trained to predict overall survival (OS) based on radiomics features derived from the primary tumor on CT images. OS was defined as the time between the primary diagnosis and death or censored at the date of last follow-up while follow-up was consequently performed for at least three years. All CT images were acquired for diagnostic purposes prior to treatment or for radiotherapy treatment planning, all with an intravenous contrast injection protocol. Feature selection was performed by fitting univariate Cox models for each individual feature, and selecting features based on their association with survival. Separate models were built using TNM8, tumor volume, clinical, and biological variables, and combinations thereof with radiomics features. Patient stratification was assessed through Kaplan-Meier (KM) curves and log-rank test for significance ( P- value<0.05). The prognostic accuracy was reported through the concordance-index (CI). Results An 11-feature radiomics signature was able to significantly stratify the validation cohort into three risk groups ( P <0.01), with a CI of 0.67. TNM8 could also determine three risk groups ( P <0.01), with a CI of 0.74 in validation. Combining radiomics, clinical and biological variables, TNM8, and volume resulted in significantly better discriminatory performance than radiomics/TNM8 alone, with a CI of 0.79 in validation (Figure 1). Figure 2 provides an overview of the CI-indices of the validation results.

Conclusion A multifactorial prognostic model for stage III and IV HNSCC (TNM7 th edition) based on simple variables available for every patient and including CT-radiomics features is able to very accurately predict OS and to significantly discern different risk-groups. The multifactorial model was found to have higher predictive performance than the current gold standard of TNM8. This could be useful in treatment (de-)escalation trials and clinical decision-support. Proffered papers: Proffered papers 40: Advances in RT planning

OC-0643 New VMAT non-coplanar solutions for mono-isocentric radiosurgery of multiple brain metastases I. Kuijper 1 , P. Van Duijvenvoorde 1 , R. Kossen 1 , B.J. Slotman 1 , J. Cuijpers 1