Scholarlink

Objectives <p>Artificial intelligence (AI) is increasingly used in dentomaxillofacial radiology, with most research focusing on image-based lesion detection and classification. Little is known about how clinicians perform when assisted by knowledge-based diagnostic tools, such as statistical models and generative large language models. This study aimed to evaluate whether AI-assisted decision support can enhance clinicians’ accuracy in diagnosing jawbone lesions.</p> Materials and methods <p>This study compared the diagnostic performance of three general dentists and three dentomaxillofacial radiologists interpreting 25 jawbone lesion cases under three conditions: unaided, assisted by ORAD (a statistical model), and assisted by RAISE (a retrieval-augmented generative AI tool). Diagnostic accuracy was compared using generalized estimating equations, and human-machine agreement was calculated.</p> Results <p>Dentists’ accuracy improved from 53% unaided to 67% with ORAD but decreased to 49% with RAISE. Dentomaxillofacial radiologists performed consistently high across the three conditions (81% unaided, 88% with ORAD, and 84% with RAISE) and significantly outperformed dentists in all conditions (<i>p</i> ≤ 0.002). Among dentists, using ORAD significantly outperformed RAISE (<i>p</i> = 0.033). Human-machine agreement was moderate for both tools.</p> Conclusions <p>In this study, ORAD support modestly improved dentists’ diagnostic accuracy, although not statistically significant, whereas generative AI did not show measurable benefit. Dentomaxillofacial radiologists’ performance was unaffected by either tool.</p> Clinical relevance <p>Generative AI tools relying solely on user input may have limited clinical utility for jawbone lesion diagnosis. Future AI systems should integrate image-based analysis with adaptive, interactive reasoning to better augment clinical decision-making. In its present form, decision-support tools seem more useful for dentists than dentomaxillofacial radiologists.</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

Impact of AI-assisted decision support on radiological diagnosis of jawbone lesions

  • 期刊论文

  • 发表日期：

    2026年1月23日

查看全文
  • Jonas Ver Berne,
  • Soroush Baseri Saadi,
  • Maria Fernanda Silva da Andrade-Bortoletto,
  • Rocharles Fontenele,
  • Reinhilde Jacobs

摘要

Objectives

Artificial intelligence (AI) is increasingly used in dentomaxillofacial radiology, with most research focusing on image-based lesion detection and classification. Little is known about how clinicians perform when assisted by knowledge-based diagnostic tools, such as statistical models and generative large language models. This study aimed to evaluate whether AI-assisted decision support can enhance clinicians’ accuracy in diagnosing jawbone lesions.

Materials and methods

This study compared the diagnostic performance of three general dentists and three dentomaxillofacial radiologists interpreting 25 jawbone lesion cases under three conditions: unaided, assisted by ORAD (a statistical model), and assisted by RAISE (a retrieval-augmented generative AI tool). Diagnostic accuracy was compared using generalized estimating equations, and human-machine agreement was calculated.

Results

Dentists’ accuracy improved from 53% unaided to 67% with ORAD but decreased to 49% with RAISE. Dentomaxillofacial radiologists performed consistently high across the three conditions (81% unaided, 88% with ORAD, and 84% with RAISE) and significantly outperformed dentists in all conditions (p ≤ 0.002). Among dentists, using ORAD significantly outperformed RAISE (p = 0.033). Human-machine agreement was moderate for both tools.

Conclusions

In this study, ORAD support modestly improved dentists’ diagnostic accuracy, although not statistically significant, whereas generative AI did not show measurable benefit. Dentomaxillofacial radiologists’ performance was unaffected by either tool.

Clinical relevance

Generative AI tools relying solely on user input may have limited clinical utility for jawbone lesion diagnosis. Future AI systems should integrate image-based analysis with adaptive, interactive reasoning to better augment clinical decision-making. In its present form, decision-support tools seem more useful for dentists than dentomaxillofacial radiologists.