Scholarlink

Objective <p>Oral diseases remain a major global health burden. Recent advances in artificial intelligence (AI) across medical imaging have encouraged similar developments in dental diagnostics. Within the spectrum of deep learning architectures, the You Only Look Once (YOLO) model has gained attention for its real-time object detection capabilities. This systematic review aims to comprehensively evaluate the scope, diagnostic performance, and methodological quality of YOLO applications in dental imaging.</p> Methods <p>A PRISMA-compliant search of PubMed, Scopus, and Google Scholar (2020–2025) identified studies applying YOLO-based detection or segmentation to dental images. Data extraction covered study characteristics, YOLO versions, datasets, annotation strategies, and performance metrics. Due to the high heterogeneity, a narrative synthesis was conducted. The risk of bias was assessed using the QUADAS-2.</p> Results <p>Seventy-three studies were included across diverse domains, including caries detection, periodontal assessment, lesion recognition, implants, and pediatric dentistry. Reported performance was generally high: F1-scores ranged from 0.63 to 0.994, and mAP50 from 0.425 to 1.0. Metrics reporting was inconsistent, as many studies provided only mAP50 rather than the more comprehensive mAP50-95 (range: 0.272–0.932), which limited comparability. Newer models (YOLOv8–YOLOv11) demonstrated improved sensitivity and multi-class detection, yet were often constrained by small, single-center datasets, reliance on augmentation, and limited external validation.</p> Conclusion <p>YOLO architectures offer strong potential as accurate and efficient diagnostic tools across dental specialties. Nonetheless, their clinical translation is hindered by dataset limitations, inconsistent reporting, and computational demands. Future research should prioritize the use of diverse datasets, standardized evaluation, and multicenter validation. Ultimately, dataset quality and clinical context matter more for performance than the YOLO version.</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

Exploring YOLO applications in dentistry through a systematic review of detection and segmentation models

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月1日

查看全文
  • Henri Hartman,
  • Rannisa Fitriasyari Amalia Ruslandi,
  • Denny Nurdin,
  • Saiful Akbar,
  • Arlette Suzy Setiawan

摘要

Objective

Oral diseases remain a major global health burden. Recent advances in artificial intelligence (AI) across medical imaging have encouraged similar developments in dental diagnostics. Within the spectrum of deep learning architectures, the You Only Look Once (YOLO) model has gained attention for its real-time object detection capabilities. This systematic review aims to comprehensively evaluate the scope, diagnostic performance, and methodological quality of YOLO applications in dental imaging.

Methods

A PRISMA-compliant search of PubMed, Scopus, and Google Scholar (2020–2025) identified studies applying YOLO-based detection or segmentation to dental images. Data extraction covered study characteristics, YOLO versions, datasets, annotation strategies, and performance metrics. Due to the high heterogeneity, a narrative synthesis was conducted. The risk of bias was assessed using the QUADAS-2.

Results

Seventy-three studies were included across diverse domains, including caries detection, periodontal assessment, lesion recognition, implants, and pediatric dentistry. Reported performance was generally high: F1-scores ranged from 0.63 to 0.994, and mAP50 from 0.425 to 1.0. Metrics reporting was inconsistent, as many studies provided only mAP50 rather than the more comprehensive mAP50-95 (range: 0.272–0.932), which limited comparability. Newer models (YOLOv8–YOLOv11) demonstrated improved sensitivity and multi-class detection, yet were often constrained by small, single-center datasets, reliance on augmentation, and limited external validation.

Conclusion

YOLO architectures offer strong potential as accurate and efficient diagnostic tools across dental specialties. Nonetheless, their clinical translation is hindered by dataset limitations, inconsistent reporting, and computational demands. Future research should prioritize the use of diverse datasets, standardized evaluation, and multicenter validation. Ultimately, dataset quality and clinical context matter more for performance than the YOLO version.