Development a software for detecting burn severity using convolutional neural network-based approach

Abstract

Burns are a significant cause of injury and can result in severe physiological reactions, metabolic disturbances, scarring, organ failure, and even death if not properly managed. Traditional clinical methods for assessing burn severity can be challenging due to various factors. In the event of a burn incident, an AI-based application can quickly analyse large amounts of data, expedite repetitive tasks like burn severity assessment, reduce subjective human errors, provide a more objective evaluation of burn severity, become more accessible in areas lacking expert medical personnel or during emergencies, and offer information-based treatment options. To address this issue, this study proposed a Deep Convolutional Neural Network (DCNN) approach to detect the severity of burn injury using real-time images of skin burns. Deep learning (DL) algorithms, namely GoogleNet, ResNet-50, and Inception-v3, were employed to train the images in Matlab software. In addition, almost 25% of the images were reserved for external validation. The developed interface achieved an accuracy rate of 90.22% in assessing burn severity based on visual data from actual cases. Consequently, by harnessing intelligent technologies, the suggested DCNN-based method can assist healthcare professionals in assessing the extent of burn injuries and delivering timely and suitable treatments. This, in turn, significantly mitigates the adverse outcomes associated with burns.