A registration-consistent longitudinal CBCT framework for quantifying treatment-induced tumour texture changes in head and neck radiotherapy

Abstract Background Head and neck squamous cell carcinoma (HNSCC) exhibits substantial inter-patient variability in treatment response. Cone-beam computed tomography (CBCT) is routinely acquired during radiotherapy but remains underutilised for longitudinal tumour characterisation. This study aimed to develop a registration-consistent longitudinal CBCT framework for quantifying treatment-induced tumour texture and to evaluate the temporal stability of radiomic features as candidate imaging biomarkers. Unlike conventional pre/post-treatment comparisons, longitudinal intra-treatment trajectories remain insufficiently characterised. Methods Thirty patients with HNSCC treated with definitive radiotherapy were retrospectively analysed. Tumour volumes defined on planning CT were propagated to daily CBCT images using rigid point-cloud registration to ensure spatial correspondence across treatment fractions. Radiomic features (entropy, strength, coarseness) were extracted within longitudinally aligned volumes of interest. Temporal dynamics were modelled using linear mixed-effects models incorporating treatment fraction, quadratic temporal terms, and tumour differentiation group (G1–G3). Results Entropy showed significant nonlinear temporal behaviour ( p < 0.001), characterised by a mid-treatment peak and substantial inter-patient variability. Strength exhibited modest nonlinear variation without significant group-dependent effects. Coarseness demonstrated the most consistent longitudinal behaviour, with significant group-dependent temporal patterns ( p < 0.01) and well-defined turning points across differentiation groups. Coarseness trajectories were less affected by high-frequency variation than entropy or strength. Conclusion Longitudinal CBCT-derived radiomic features capture dynamic tumour changes during radiotherapy, but their robustness varies substantially. Coarseness emerged as the most stable and interpretable feature, and the most promising candidate descriptor, although prospective validation with clinical endpoints is required before any biomarker claims can be made. The proposed framework provides a methodological basis for developing CBCT-based imaging biomarkers for longitudinal treatment monitoring and adaptive radiotherapy.