牙外伤Evo——开发支持牙外伤专业管理的人工智能聊天机器人
本研究基于国际牙外伤协会指南开发了规则驱动的人工智能聊天机器人Dental Trauma Evo,由牙髓病和儿童口腔专科医生验证,并通过384次涵盖32种外伤类型的交互评估性能。结果显示推荐临床管理的准确率达100%,但初版对乳牙釉质、牙本质和牙髓暴露骨折的治疗选项完整性仅53%,代码调整后多数情况完整性达100%,仅开放根尖撕脱牙再植建议为93%。
要点
- 基于国际牙外伤协会指南的规则驱动聊天机器人可提供与指南一致的牙外伤管理建议。
- 384次交互评估中推荐临床管理准确率为100%。
- 初版对乳牙釉质、牙本质和牙髓暴露骨折的回复完整性仅53%,调整后多数达100%,开放根尖撕脱牙再植建议为93%。
核心结论
- 人工智能聊天机器人在标准化牙外伤管理方面表现出高性能,证明其有潜力通过提供与国际指南一致的快速精准回复来辅助专业人员决策。
研究设计与样本
- 基于国际牙外伤协会指南、采用规则驱动系统开发Dental Trauma Evo聊天机器人,用Python实现并集成ChatGPT-4 API,通过Streamlit提供50多种语言交互。
- 由牙髓病和儿童口腔专科医生评估回复的清晰度和一致性,并通过384次涵盖32种外伤类型的交互评估推荐的准确性和完整性。
主要发现
- 聊天机器人推荐适当临床管理的准确率达100%。
- 部分初始回复未涵盖指南列出的所有治疗选项,尤其是乳牙釉质、牙本质和牙髓暴露骨折,仅53%的回复完整。
- 代码调整后第二轮测试显示显著改善,多数情况完整性达100%。
- 仅开放根尖撕脱牙再植建议的完整性为93%,提示需要进一步改进。
局限性
- 部分初始回复未涵盖指南列出的所有治疗选项,尤其是乳牙釉质、牙本质和牙髓暴露骨折,仅53%的回复完整。
- 开放根尖撕脱牙再植建议的完整性为93%,提示需要进一步改进。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
INTRODUCTION: Dental trauma is a public health issue that requires proper clinical management to minimize complications. Artificial intelligence-based technologies can support decision-making by providing standardized guidance. This study developed and evaluated an artificial intelligence-powered chatbot to assist professionals in managing dental trauma. METHODS: The Dental Trauma Evo chatbot was developed based on the guidelines of the International Association of Dental Traumatology, utilizing a rule-based system to ensure recommendations aligned with established clinical protocols. The chatbot was implemented in Python, integrated with the ChatGPT-4 API, and made available via Streamlit, enabling interactions in over 50 languages. Validation was conducted by specialists in endodontics and pediatric dentistry, who assessed the clarity and consistency of the responses. Performance was evaluated through 384 interactions covering 32 types of trauma, assessing the accuracy and completeness of the recommendations. RESULTS: The chatbot achieved 100% accuracy in recommending appropriate clinical management. However, some initial responses did not include all treatment options outlined in the guidelines, particularly in cases of enamel, dentin, and pulp-exposed fractures in primary teeth, where only 53% of responses were complete. After adjustments to the code, the second round of testing showed significant improvements, reaching 100% completeness in most cases. Only the recommendations for the reimplantation of avulsed teeth with an open apex reached 93%, indicating the need for further refinements. CONCLUSIONS: The artificial intelligence-powered chatbot demonstrated high performance in standardizing the management of dental trauma, proving its potential to assist professionals in decision-making by providing quick and precise responses aligned with international guidelines.
引用
- Bubna DP, Felipe de Jesus Freitas P, Ferraz AX, et al. Dental Trauma Evo - Development of an Artificial Intelligence-powered Chatbot to Support Professional Management of Dental Trauma. J Endod. 2025. PMID: 40449881. doi:10.1016/j.joen.2025.05.012
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。