Feature-based vs. deep-learning fusion methods for the in vivo detection of early radiation dermatitis using Optical Coherence Tomography

Abstract

Abstract More than half of all cancer patients receive some form of radiation therapy during the course of their illness. Unfortunately, acute radiation dermatitis (ARD) is a common side effect of radiation that leads to significant morbidity. Although there are various treatment options, ARD is still the cause of significant distress, thus, additional research is required to improve prevention and treatment strategies. Unfortunately, the lack of biomarkers for quantitative assessment of early changes associated with the condition, impedes further progress. This study was designed to explore the identification of early ARD using intensity-based and novel features of Optical Coherence Tomography (OCT) images, combined with machine learning. Twenty-two patients underwent imaging twice weekly, at six locations on the neck, until the end of their radiation treatment. An expert oncologist graded the severity of their ARD. A traditional feature-based machine learning (ML) and a deep learning (DL) fusion approach were compared for their ability to classify normal skin vs. early ARD from the 1487-image dataset collected. Results showed that the deep learning approach outperformed traditional ML, achieving an accuracy of 88%. These findings provide a promising foundation for future research aimed at creating a quantitative assessment tool to improve the management of ARD.