S586
Clinical – Head & neck
ESTRO 2026
Electrical and Computer Engineering, University of Cyprus, Nicosia, Cyprus
(UHZ and MUV) or 3T MRI (CLB and GR), performed in treatment position using treatmentimmobilization masks. GTV delineation as well as intra-OV were performed on both modalities. Three additional RO were involved to evaluate inter-OV. Several weeks elapsed between sessions to minimize recall bias. Volumetric, Dice Similarity Coefficient (DSC) and Hausdorff values (HD95) analyses were calculated. Statistical significance (p<0.05) was assessed by pair- wise Wilcoxon tests. Results: Qualitatively, delineations were easierand faster on MRI T1G images (40 min vs 60 min) compared to ceCT. Quantitatively, for primary tumors, MRI showed significantly larger GTVtvolumes (median difference of +2.1 cm ³ , p = 0.012). This differences were most pronounced for oropharyngeal tumors. Spatial agreement between ceCT and MRI was highest with T1G (mean DSC = 0.737 ± 0.151 ; HD95 = 3.25 ± 1.60 mm). No added value of T1- and T2-sequences was observed. Inter-OV was comparable for tumors between ceCT and MRI T1G (CT: DSC = 0.82 ± 0.09; T1G: 0.84 ± 0.07; p = 0.24) with comparable geometric precision (HD95: 2.82 ± 2.63 mm for ceCT vs 2.68 ± 2.48 mm for T1G). Intra-OV was not different between ceCT and MRI. For nodal GTVs, no pair-wise difference existed between ceCT and T1G. No significant inter- observer differences were found for the delineated volume. Conclusion: Planning T1G MRI demonstrated both qualitative and quantitative difference compared to ceCT especially for the delineation of oropharyngeal carcinoma by improving consistency, inter-OV variability, and enhanced geometric agreement suggesting its integration for GTV definition. No added value of MRI was observed for nodal delineation. The full analysis for the 190 patients will be presented at ESTRO. Keywords: head and neck, gross tumor volume delineation, MRI Comparative Evaluation of Feature-Based and Deep Learning Approaches Using Advanced Imaging for Early Detection of Radiation Dermatitis Iosif Strouthos 1,2 , Constantina Cloconi 1 , Georgios Antorkas 3 , Yiannis Roussakis 3 , Melka Benjamin 1 , Christos Photiou 4 1 Radiation Oncology, German Oncology Center, Limassol, Cyprus. 2 Clinical Faculty, European University Cyprus, Nicosia, Cyprus. 3 Medical Physics, German Oncology Center, Limassol, Cyprus. 4 KIOS Research and Innovation Center of Excellence, Department of Digital Poster 2105
Purpose/Objective: Radiation-induced acute radiation dermatitis (ARD) is a common side effect of cancer radiotherapy, causing discomfort and reducing patients’ quality of life. Existing management strategies are hampered by the absence of dependable biomarkers for early detection and severity evaluation. This study investigates the potential of combining Optical Coherence Tomography (OCT) imaging with machine learning techniques to identify ARD at an early stage, contrasting traditional feature-based classifiers with modern deep learning models. Material/Methods: The study involved 22 patients undergoing radiation therapy for head and neck cancer. OCT scans were performed twice weekly at six designated neck locations throughout treatment. The severity of ARD was clinically rated by experienced oncologists and correlated with OCT imaging data. Two machine learning approaches were employed: a conventional feature-based classifier and a deep learning model utilizing late fusion. A total of 1,487 OCT images were analyzed to distinguish healthy skin from areas affected by ARD. Results: Findings revealed that the deep learning model significantly outperformed the traditional feature- based approach, achieving an accuracy of 88%. This indicates that the deep learning method is more effective in differentiating normal from ARD-affected skin based on OCT images. The deep learning approach demonstrated greater robustness in managing the variability and complexity inherent in clinical imaging data. Conclusion: These results suggest that deep learning models integrated with OCT imaging hold significant promise for early ARD detection and monitoring. This combination offers an objective, quantitative tool for clinical assessment, which could facilitate more personalized and timely treatment strategies for patients with radiation dermatitis. Further studies involving larger patient populations are necessary to validate these findings and explore the integration of such techniques into routine clinical workflows. Overall, this research marks progress toward developing reliable, non-invasive diagnostic tools aimed at improving oncology patient care. Keywords: Optical Coherence Tomography, Radiation Dermatitis
Made with FlippingBook - Share PDF online