Evaluation of the Performance of Artificial Intelligence Based Chatbots in Providing First Aid Information on Dental Trauma According to the ToothSOS Application

Abstract

AIM: The aim of this study was to evaluate the performance of ChatGPT-4o and Gemini Advanced artificial intelligence-based chatbots (AI-based chatbots) in providing emergency intervention recommendations for dental trauma with intraoral photographs of patients diagnosed with traumatic dental injuries, and to assess their compatibility with emergency intervention recommendations in the ToothSOS application. MATERIAL AND METHODS: In this study, 80 intraoral photographs obtained from patients presenting with dental trauma were uploaded to two different AI-based chatbots (ChatGPT-4o and Gemini Advanced) and the responses generated by these systems were evaluated by four paediatric dentists. The evaluators scored the responses with a Modified Global Quality Score (GQS), referring to the English instructions of the ToothSOS application. In order to analyse the reliability of the responses, a total of three evaluation sessions were conducted 1 week apart. RESULTS: The ChatGPT-4o performed better when all injury types were considered together (p = 0.012). It was found that the ChatGPT-4o performed much better in complicated crown fracture cases (p = 0.004) and the Gemini Advanced chatbot performed much better in critical dental injuries such as avulsion (p < 0.001). CONCLUSIONS: AI-based chatbots can be a helpful tool in the assessment of dental trauma. However, further development and expert validation are needed to improve their accuracy and consistency, especially in complex cases. Incorporating the International Association of Dental Traumatology (IADT) guidelines into the databases of these systems could improve the reliability of their recommendations. In addition, given the widespread use of AI-based chatbots in many fields, particularly health, they could contribute to public health by supporting access to accurate information.

Ähnliche Arbeiten