Evidence-Based Potential of Generative Artificial Intelligence Large Language Models on Dental Avulsion: ChatGPT...

Area of Science:

• Dental Traumatology
• Artificial Intelligence in Healthcare
• Natural Language Processing

Background:

• Dental avulsion, a common dental emergency, requires accurate information for effective management.
• Artificial intelligence (AI) language models offer potential for information dissemination in healthcare.
• Evaluating the accuracy of AI responses in specialized fields like dentistry is crucial.

Purpose of the Study:

• To comparatively evaluate the accuracy and comprehensiveness of answers provided by ChatGPT and Gemini regarding dental avulsion.
• To assess the performance of AI language models against established dental traumatology guidelines.

Main Methods:

• Thirty-three questions on dental avulsion were formulated based on International Society of Dental Traumatology (IADT) guidelines.
• Questions included multiple-choice, binary, and open-ended formats, covering technical and patient queries.
• Responses from ChatGPT and Gemini were scored by four pediatric dentists, with statistical analyses including ICC and Mann-Whitney U tests.

Main Results:

• Gemini achieved a statistically significantly higher mean score than ChatGPT (p=0.001).
• While ChatGPT excelled in open-ended and true/false questions, it showed lower accuracy in multiple-choice questions.
• Gemini's responses showed no significant difference in accuracy across different question types (p=0.088), and overall, Gemini answers were significantly more accurate (p=0.004).

Conclusions:

• Both Gemini and ChatGPT show potential for providing information on dental avulsion, guided by IADT standards.
• Further research, clinical validation, and model improvements are essential for successful integration of these large language models (LLMs) into clinical practice.