Scholarlink

Background <p>This scoping review evaluates the current state of generative artificial intelligence (AI) in implant dentistry, focusing on the performance, clinical applications, and inherent limitations of large language models (LLMs) in both clinical and educational settings.</p> Methods <p>A scoping review was conducted in accordance with PRISMA-ScR guidelines. A comprehensive search across four electronic databases (PubMed, Web of Science, Scopus, and Embase) was performed for literature published through December 2025. Eighteen eligible studies were analyzed to assess model architectures, specific task performance, and comparative proficiency against human experts.</p> Results <p>The included studies, all published in 2024 or 2025, were classified as examining patient interaction (<i>n</i> = 9), medical knowledge (<i>n</i> = 7), or diagnostic assessment (<i>n</i> = 2). The analysis revealed a distinct evolution in performance with the advent of advanced reasoning models (e.g., ChatGPT-o1, DeepSeek-R1), which occasionally surpassed licensed dentists in certification examinations. Comparative assessments between medical-specific models and general-purpose LLMs yielded divergent outcomes, indicating that domain specialization does not inherently guarantee superior clinical accuracy against state-of-the-art generalist architectures. Nevertheless, reliability remains a concern; despite the integration of retrieval-augmented generation, hallucinations persist—especially in systematic search tasks—and the inability of text-based models to interpret diagnostic imaging continues to limit their clinical autonomy.</p> Conclusions <p>Although generative AI has attained expert-level proficiency in theoretical knowledge retrieval, it currently serves best as an adjunctive support system, rather than a replacement for clinical judgment. Given the persistent risks of hallucination and the lack of visual processing capabilities, strict professional oversight is mandatory. Future research must prioritize the development of multimodal models and validate clinical outcomes through randomized trials.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Large language models in implant dentistry: a scoping review of applications, performance, and limitations

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月30日

查看全文
  • Hyun-Jun Kong,
  • Yu-Lee Kim

摘要

Background

This scoping review evaluates the current state of generative artificial intelligence (AI) in implant dentistry, focusing on the performance, clinical applications, and inherent limitations of large language models (LLMs) in both clinical and educational settings.

Methods

A scoping review was conducted in accordance with PRISMA-ScR guidelines. A comprehensive search across four electronic databases (PubMed, Web of Science, Scopus, and Embase) was performed for literature published through December 2025. Eighteen eligible studies were analyzed to assess model architectures, specific task performance, and comparative proficiency against human experts.

Results

The included studies, all published in 2024 or 2025, were classified as examining patient interaction (n = 9), medical knowledge (n = 7), or diagnostic assessment (n = 2). The analysis revealed a distinct evolution in performance with the advent of advanced reasoning models (e.g., ChatGPT-o1, DeepSeek-R1), which occasionally surpassed licensed dentists in certification examinations. Comparative assessments between medical-specific models and general-purpose LLMs yielded divergent outcomes, indicating that domain specialization does not inherently guarantee superior clinical accuracy against state-of-the-art generalist architectures. Nevertheless, reliability remains a concern; despite the integration of retrieval-augmented generation, hallucinations persist—especially in systematic search tasks—and the inability of text-based models to interpret diagnostic imaging continues to limit their clinical autonomy.

Conclusions

Although generative AI has attained expert-level proficiency in theoretical knowledge retrieval, it currently serves best as an adjunctive support system, rather than a replacement for clinical judgment. Given the persistent risks of hallucination and the lack of visual processing capabilities, strict professional oversight is mandatory. Future research must prioritize the development of multimodal models and validate clinical outcomes through randomized trials.