Abstract
INTRODUCTION: Burn injuries are a significant challenge in clinical and military settings, requiring accurate and timely assessment to guide treatment. Traditional methods for determining burn depth, a key factor in severity, rely heavily on subjective evaluation, leading to variability and delays in decision-making. Advances in Artificial Intelligence (AI) offer solutions to improve diagnostic accuracy and standardization. This study aims to evaluate the diagnostic performance of an AI model for burn depth assessment by comparing its outputs against a gold standard-focusing on image-based diagnosis of burn type and depth. MATERIALS AND METHODS: This study analyzed 29 burn patients, under an Institutional Review Board-approved protocol (IRB# 12,689) at the Eskenazi Burn Center, Indianapolis. Digital images of burns were collected and classified into 3 burn depth categories: first-degree, second-degree, and third-degree. The AI model was fine-tuned on 131 annotated digital images, augmented to 1,200 using techniques such as rotation, flipping, and brightness adjustment. Style transfer using a machine learning models (called GAN) was used to further enhance the dataset by simulating burn variations. Zero-shot (meaning no previous training) segmentation, employing pretrained foundation models, was used to localize burn regions without task-specific training. RESULTS: The proposed AI prediction model achieved 79% accuracy in classifying 3 burn depth categories. Data augmentation improved performance, while segmentation demonstrated strong utility, particularly in identifying burn regions effectively in diverse scenarios. Style transfer augmented the dataset by simulating realistic burn appearances, further enhancing model robustness. Zero-shot segmentation, meaning it identified burn areas without any prior training on similar images, successfully localized burn regions, aligning with clinical expectations. CONCLUSIONS: This study highlights the potential of AI in improving burn depth classification and segmentation. The results demonstrate that integrating AI-driven models into clinical care can enhance diagnostic accuracy, efficiency, and scalability, offering transformative tools for clinical and military applications in burn care. These methods provide a foundation for automated and standardized burn assessment, improving outcomes across diverse settings.