ESTRO 2025 - Abstract Book

S1120

Clinical – Head & neck

ESTRO 2025

of Oral Oncology, The University of Texas MD Anderson Cancer Center, Houston, USA. 4 Department of Head and Neck Surgery, The University of Texas MD Anderson Cancer Center, Houston, USA

Purpose/Objective: Osteoradionecrosis of the jaws (ORNJ) is a debilitating complication and affects up to 15% of head and neck cancer (HNC) patients who undergo radiotherapy (RT) 1 . Early and accurate detection of ORNJ is essential in improving patient outcomes and quality of life. The ClinRad staging system 2 (recommended in the latest ASCO guidelines 3 ) incorporates radiographic features for classifying ORN severity, (i.e., vertical extent of bone involvement). Variability in imaging modalities and specialty-specific knowledge contribute to current disparities in diagnosing and classifying ORNJ. This study aims at benchmarking physician performance for diagnosing and staging radiographic ORNJ(rORNJ). Material/Methods: A prospective inter-rater agreement study was conducted involving 20 healthcare providers across varying specialties including dentistry, radiation oncology, surgery, and neuroradiology, at The University of Texas with duplicates (n=5) for assessment of intra-observer variability. Each image modality was reviewed independently and as paired sets using the ASCO/ISOO/MASCC-endorsed ClinRad ORN grading system. Diagnostic performance, sensitivity, and specificity were assessed using ROC analysis, and intra- and inter-observer agreement measured with Fleiss’ Kappa. Analysis also considered physician specialty and years of clinical experience impact on early detection. Results: The overall AUC for all grouped observers was 0.65 (CI:0.59-0.71) for non-paired CT-only assessments and increased to 0.85 (CI: 0.81, 0.91) with paired CT-OPG imaging. Paired 2D-3D imaging improved diagnostic performance, with AUC values ranging from 0.79 (CI: 0.68-0.83) for residents to 0.98 (CI: 0.94-0.99) for surgeons (Figure 1). Variability in sensitivity and specificity was observed across specialties; dentists with 10–15 years of experience achieved the highest accuracy (61.3%, CI: 0.49-0.73) using single-modality imaging. Inter-rater agreement for ORN detection was limited, with a median Fleiss Kappa value of 0.06. Misclassification of non-ORN cases as Stage I was most prevalent among residents (72.1%). The most commonly reported radiographic features for confirmed ORN cases staged as ClinRad 1 or 2 included “bone necrosis involving the basilar bone or maxillary sinus”, "bone necrosis confined to alveolar bone" and "bone lysis/sclerosis." This study uniquely provides novel benchmarking for physician detection/grading of rORNJ. MD Anderson Cancer Center. Participants reviewed 85 anonymized, paired imaging sets (3D CT and 2D orthopantomogram (OPG)) from 30 patients with confirmed ORN,

Conclusion: This study establishes an essential benchmark for physician detection of rORNJ that will enable performance comparison with computational approaches. The significant variability in diagnostic