Sie befinden Sich nicht im Netzwerk der Universität Paderborn. Der Zugriff auf elektronische Ressourcen ist gegebenenfalls nur via VPN oder Shibboleth (DFN-AAI) möglich. mehr Informationen...
Visual inspection, along with physical examination, is the traditional method to assess burns. However, burn-care providers have different levels of experience and may face challenges in assessing the depth and severity of the wounds. The challenges associated with the traditional approach, such as poor and varying diagnosis/prognosis accuracy, have inspired researchers towards automated burn assessment to ensure effective burn wound management. The current research aims to improve automatic burn wound assessment. It provides an ordered scoring scale to measure burn severity using four characteristics: inflammation, scar, uniformity, and pigmentation. The research also proposes an attention-based Convolutional Neural Network (CNN) model to assess the characteristics of burn wounds. The model is evaluated with 2D color images to assess levels of inflammation, scar, uniformity, and pigmentation with two different datasets, and the performances are compared with other models. The attention mechanism of the deep learning model selectively focuses on salient parts of the image to improve the understanding of the visual structure and enhance the classification accuracy. The proposed work outperforms most prior related work, achieving 93% in average accuracy.Clinical relevance-This research has significant clinical relevance in assisting accurate, reliable, and on-time diagnosis, treatment, and follow-up of burn wounds and thereby, provides effective burn wound management.