Automated Surgical Wound Classification for Intelligent Emergency Care Applications

Abstract

Developing AI-powered solutions for emergency care is critical to improving patient outcomes. This study presents an advanced AI model designed to accurately classify surgical wounds to facilitate prompt and appropriate emergency response. We used two state-of-the-art image classification models, ResNeXt and Vision Transformer (ViT), to evaluate their effectiveness in classifying wound images. These models were selected based on preliminary evaluations of several popular models, and their superior performance metrics justified their final selection. The models were trained on a combined dataset of approximately 1,000 images from the Medetec and AZH(Advancing the Zenith of Healthcare) datasets. To improve classification accuracy, the dataset was preprocessed, including image resizing, flipping, normalization, and augmentation. The ViT model demonstrated superior performance, achieving an accuracy of 92.78%, precision of 94.89%, recall of 91.87%, and an F1 score of 92.44%. These results surpass those of existing studies such as Shenoy et al. (85.1% accuracy), Rostami et al. (68.7% accuracy), and Gao et al. (76.6% accuracy). Our proposed AI system not only accelerates life-saving first aid by providing timely and accurate wound classification but also enhances the skills of emergency responders through continuous learning and real-time feedback. By integrating AI into emergency service protocols, we aim to improve response times and collaboration among medical personnel, ultimately contributing to better patient survival rates.