Development of an explainable AI-based disaster casualty triage system

Disaster response and management are critical components of rescue team training in countries worldwide. In addition to conducting various disaster drills, rescue teams are trained to perform casualty triage in simulated scenarios, allowing medical personnel to provide optimal treatment based on triage classifications. Due to the necessity of adapting disaster scenarios to enable responders to handle diverse disaster sites, each scenario must be interactive, helping rescuers understand how to perform triage effectively during disaster response. To enrich the variety of scenarios, AI can now be utilized for scenario design. However, for more rational script creation, generative AI must be grounded in Explainable AI (XAI) to make the generation process transparent, thus enhancing the scenario?s realism. This paper proposes an XAI-based disaster casualty triage scenario system. The system generates scenarios through generative AI, utilizing XAI to ensure data transparency. The primary output is a simulation training game focused on disaster scenarios, developed on the Unity platform to build realistic accident scenes. The goal is to provide frontline firefighters with immersive training to strengthen their on-site response and emergency handling skills. The game incorporates a triage mechanism that guides users to categorize injuries based on symptoms and apply appropriate medical actions, aiming to minimize casualties during disasters. From an educational perspective, this game provides the general public with an understanding of how firefighters perform triage based on injury symptoms in emergencies, ensuring that each casualty receives necessary medical support within the golden rescue window. Through simulation and decision-making training in the game, users enhance their judgment and responsiveness, further improving their rapid reaction and handling skills in disaster scenarios.