ESTRO 2026 - Abstract Book PART I

S593

Clinical – Head & neck

ESTRO 2026

Purpose/Objective: Osteoradionecrosis (ORN) of the mandible is a debilitating and potentially life-threatening late complication of head and neck radiation therapy (RT). While the diagnosis of ORN is usually straightforward, a subset of patients may present with tumor recurrence developing within or adjacent to necrotic bone areas. Differentiating these entities using conventional imaging remains challenging, even with contrast-enhanced CT or FDG PET/CT. Radiomic analysis may help quantify subtle differences in tissue morphology and texture that are not perceptible by visual assessment. This exploratory study aimed to identify CT-based radiomic features associated with tumor recurrence among patients with pathologically confirmed ORN. Material/Methods: This retrospective study included 24 patients who underwent pathologically proven mandibular ORN surgery between 2019 and 2024, all of whom had preoperative contrast-enhanced CT. The ORN volume was manually delineated under the supervision of an experienced head-and-neck radiologist. Intensity discretization was performed using a fixed bin width of 25 HU, after resampling to 1 × 1 × 1 mm ³ , without absolute rescaling, in accordance with IBSI recommendations for CT radiomics. A total of 107 shape, histogram, and texture features (GLCM, GLRLM, GLSZM, NGTDM) were extracted using a standardized Python workflow compliant with IBSI recommendations. Univariate comparisons between pure ORN and ORN with recurrence (ORN+Recurrence) were performed using the Wilcoxon rank-sum test. Significant features (p < 0.05) were tested for inter- feature redundancy with Spearman’s correlation ( ρ < 0.7). Robustness was assessed using 1000 bootstrap resamplings. Results: Among 24 patients (20 male, median age 69.5 years), 21 presented with pure ORN and 3 with ORN+Recurrence. The median interval from RT to ORN onset was 15.4 months, and 62.5% had received concurrent chemotherapy. Three features were significantly different between groups: Least Axis Length (p = 0.041), Kurtosis (p = 0.031), and Contrast from neighborhood gray tone difference matrix (NGTDM) (p = 0.023) (Figure 1). Recurrent lesions exhibited larger minimal axis dimension, lower intensity kurtosis, and higher texture contrast. Correlations between features remained below ρ = 0.7 (Figure 2), and bootstrap analysis confirmed the consistency but limited robustness of these associations due to the small number of recurrence cases (n = 3).

Conclusion: There was no indication of unintended underdosage in the accumulated dose at recurrence positions. These findings show that current PTV margins and high- quality image guided photon therapy procedures are sufficient to avoid unintended cold spots of clinical consequence in oropharyngeal cancer. References: (1) Håkansson, K. et al. A comparison of loco-regional relapse pattern in HPV positive vs negative oropharyngeal cancers following radiotherapy; relation to pretreatment FDG-PET and radiotherapy target volumes. Acta Oncol 62, 1028–1035 (2023).(2) Due, A. K. et al. Methodologies for localizing loco- regional hypopharyngeal carcinoma recurrences in relation to FDG-PET positive and clinical radiation therapy target volumes. Acta Oncol 49, 984–990 (2010). Keywords: Dose accumulation, recurrence pattern Radiomic analysis of pathologically confirmed mandibular osteoradionecrosis to identify cases with associated recurrence: a retrospective study Olivier OZCAN 1 , Remy LALUC 1 , Aline CARSIN-VU 2 , Nathaniel ASSOULY 3 , Esteban BRENET 4 , Antonio DA SILVA RIBEIRO MOTA 1 , Stéphane DERRUAU 5 , Marine FONTAINE 1 , Camille INVERNIZZI 1 , Oriane MARQUES 6 , Nicolas PASSAT 7 , Arnaud BEDDOK 1 1 Radiation oncology, Insitut Godinot, Reims, France. 2 Radiology, CHU de Reims, Reims, France. 3 ENT, Insitut Godinot, Reims, France. 4 ENT, CHU de Reims, Reims, France. 5 Dentistry, CHU de Reims, Reims, France. 6 Radiation oncology, Centre ICONE, Bezannes, France. 7 CRESTIC, CRESTIC, Reims, France Digital Poster 2278

Made with FlippingBook - Share PDF online