ESTRO 2025 - Abstract Book

S3910

Radiobiology - Normal tissue radiobiology

ESTRO 2025

439

Digital Poster Advanced Imaging and AI Techniques for Early Detection of Radiation Dermatitis: Feature-Based vs. Deep Learning Models Iosif Strouthos 1,2 , Christos Photiou 3 , Yiannis Roussakis 4 , Melka Benjamin 1 , Constantina Cloconi 1 , Konstantinos Ferentinos 1,2 1 Radiation Oncology, German Oncology Center, Limassol, Cyprus. 2 Medical Faculty, European University Cyprus, Nicosia, Cyprus. 3 KIOS Research and Innovation Center of Excellence, University of Cyprus, Nicosia, Cyprus. 4 Medical Physics, German Oncology Center, Limassol, Cyprus Purpose/Objective: Acute radiation dermatitis (ARD) is a prevalent side effect of radiation therapy, leading to discomfort and impacting the quality of life of cancer patients. Current approaches to managing ARD are limited by the lack of reliable biomarkers for early detection and severity assessment. This study aims to explore the efficacy of Optical Coherence Tomography (OCT) imaging combined with machine learning for the early detection of ARD, comparing traditional feature-based methods with advanced deep learning techniques. Material/Methods: Twenty-two patients undergoing radiation therapy for head and neck cancer participated in the study. OCT imaging was performed twice weekly across six predetermined neck locations for each patient throughout the course of their treatment. The severity of ARD was clinically assessed by an experienced oncologist and correlated with the OCT images. Two machine learning approaches were utilized: a traditional feature-based classifier and a deep learning model employing a late-fusion technique. A total of 1487 OCT images were processed to classify healthy skin versus ARD-affected areas. Results: The study found that the deep learning model significantly outperformed the traditional machine learning approach. The deep learning method achieved an accuracy rate of 88%, indicating its superior ability to differentiate between normal and affected skin tissues based on OCT image analysis. This approach proved to be more robust in handling the variability and complexity of clinical imaging data compared to the traditional feature-based method. Conclusion: The findings suggest that deep learning models, when integrated with OCT imaging, offer a promising tool for the early detection and monitoring of ARD. By providing a quantitative and objective assessment method, this approach can enhance clinical decision-making, potentially leading to more personalized management strategies for patients experiencing radiation dermatitis. Further research is encouraged to validate these findings across a larger patient cohort and explore the integration of these techniques into routine clinical practice. These results demonstrate a step forward in developing reliable, non-invasive diagnostic tools to improve patient care in oncology settings.

Keywords: Radiation dermatitis, Optical coherence tomography

References: https://link.springer.com/article/10.1007/s10278-024-01241-4 https://opg.optica.org/abstract.cfm?uri=OCT-2022-CM2E.4