KEY POINTS
- This multicentre retrospective study included 24 patients with pathologically confirmed mandibular osteoradionecrosis who underwent surgery and had preoperative computed tomography available.
- Patients were classified by surgical pathology as pure osteoradionecrosis (n = 21) or osteoradionecrosis with associated tumour recurrence (n = 3). Isolated mandibular recurrence without histological osteoradionecrosis was excluded.
- A total of 107 radiomic features were extracted using an Image Biomarker Standardization Initiative-compliant pipeline after manual lesion segmentation and radiologist review.
- Three features showed nominal uncorrected differences: least axis length (P = .041), kurtosis (P = .031), and neighbouring gray-tone difference matrix contrast (P = .023). None remained significant after false-discovery-rate correction; adjusted P = .535 for all three.
- The direction of association was consistent across the three recurrence cases and in leave-one-out analysis, but no multivariable model or classifier was attempted because of the very small number of recurrence events.
CLINICAL TAKEAWAY
Computed tomography radiomics may capture subtle differences between pure mandibular osteoradionecrosis and osteoradionecrosis harbouring tumour recurrence. However, with only three recurrence cases and no corrected statistically significant features, this is strictly hypothesis-generating and should not guide biopsy, surgery, or oncologic decision-making without external validation.