Durmazpinar Parla Meva, Ekmekci Ece
Department of Endodontics, Faculty of Dentistry, Marmara University, İstanbul, Turkey.
BMC Oral Health. 2025 Mar 29;25(1):457. doi: 10.1186/s12903-025-05857-y.
Artificial intelligence (AI) technologies have revolutionized fields such as economics, law, and healthcare. Large language models, like ChatGPT, have shown significant potential in dentistry, supporting diagnostic accuracy, treatment planning, and education. However, earlier versions of ChatGPT were limited to text-based data. The latest multimodal model, ChatGPT-4o, introduced in 2024, now processes text, images, audio, and video, enabling broader applications in clinical education. This study evaluates ChatGPT-4o's diagnostic accuracy in endodontic cases, comparing it with dental students' performance.
This study included two groups of dental students: 3rd-year and 5th-year, alongside ChatGPT-4o. Participants answered 15 multiple-choice questions designed using radiographs, clinical photographs, and patient histories. These questions, based on the American Association of Endodontists' Clinical Guidelines, were administered via Google Forms for students and ChatGPT-4o. Responses were categorized as correct, incorrect, or unanswered. Data were analyzed statistically.
ChatGPT-4o demonstrated a higher accuracy rate and lower error rate compared to students, with 91.4% correct and 8.2% incorrect responses. Third-year students had a correct response rate of 60.8%, while fifth-year students achieved 79.5%. A statistically significant difference was found between the study groups in terms of correct response rates (p < 0.05), with ChatGPT outperforming both student groups (p < 0.001). Additionally, fifth-year students showed a higher correct response rate compared to third-year students.
ChatGPT-4o demonstrates significant potential as a diagnostic support tool in dental education, particularly in endodontics. Its high diagnostic accuracy and consistency highlight its value as an innovative ( application in clinical training and decision-making.
人工智能（AI）技术已经给经济学、法律和医疗保健等领域带来了变革。像ChatGPT这样的大语言模型在牙科领域显示出了巨大潜力，有助于提高诊断准确性、辅助治疗计划制定以及支持教育工作。然而，早期版本的ChatGPT仅限于基于文本的数据。2024年推出的最新多模态模型ChatGPT-4o现在能够处理文本、图像、音频和视频，从而在临床教育中有更广泛的应用。本研究评估ChatGPT-4o在牙髓病病例中的诊断准确性，并将其与牙科学生的表现进行比较。
本研究包括两组牙科学生：三年级和五年级学生，以及ChatGPT-4o。参与者回答了15道基于X光片、临床照片和患者病史设计的多项选择题。这些问题基于美国牙髓病学家协会的临床指南，通过谷歌表单对学生和ChatGPT-4o进行测试。答案分为正确、错误或未回答。对数据进行统计学分析。
与学生相比，ChatGPT-4o显示出更高的准确率和更低的错误率，正确回答率为91.4%，错误回答率为8.2%。三年级学生的正确回答率为60.8%，而五年级学生为79.5%。在正确回答率方面，研究组之间存在统计学上的显著差异（p < 0.05），ChatGPT的表现优于两个学生组（p < 0.001）。此外，五年级学生的正确回答率高于三年级学生。
ChatGPT-4o作为牙科教育中的诊断支持工具具有巨大潜力，特别是在牙髓病学方面。其高诊断准确性和一致性凸显了它在临床培训和决策中的创新应用价值。
