Scholarlink

<p>Dental implant classification plays a vital role in treatment planning and postoperative care, yet traditional methods remain subjective and error-prone. While deep learning has advanced medical image analysis, limited open-access datasets and insufficient benchmarking of state-of-the-art models present a critical research gap. The problem addressed in this work is lack of systematic evaluation of the latest YOLO variants for accurate dental implant classification using publicly available data. The objectives were to compare YOLOv8, YOLOv9, and YOLOv10 in terms of accuracy, precision, recall, F-score, and AUC-ROC, thereby assessing their applicability in real-world clinical scenarios. The methodology considered an open-access dataset of 5,273 images across four classes, endosteal, subperiosteal, transosteal, and zygomatic, followed by hyperparameter optimization and training of the three YOLO models. Findings revealed a clear progression in model performance, with YOLOv8 achieving 93.2% accuracy and AUC-ROC of 0.94, YOLOv9 achieving 95.4% accuracy and AUC-ROC of 0.96, and YOLOv10 achieving 97.1% accuracy and AUC-ROC of 0.98. In conclusion, YOLOv10 outperformed its predecessors, establishing itself as a reliable solution for dental implant classification, thereby addressing the identified research gap.</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

Advancing dental implant classification through YOLO-based deep learning models

  • 期刊论文

  • 发表日期：

    2025年11月21日

查看全文
  • Ashwini D. Khairkar,
  • Sonali Kadam,
  • Pankaj Kadam,
  • Sujit Deshpande

摘要

Dental implant classification plays a vital role in treatment planning and postoperative care, yet traditional methods remain subjective and error-prone. While deep learning has advanced medical image analysis, limited open-access datasets and insufficient benchmarking of state-of-the-art models present a critical research gap. The problem addressed in this work is lack of systematic evaluation of the latest YOLO variants for accurate dental implant classification using publicly available data. The objectives were to compare YOLOv8, YOLOv9, and YOLOv10 in terms of accuracy, precision, recall, F-score, and AUC-ROC, thereby assessing their applicability in real-world clinical scenarios. The methodology considered an open-access dataset of 5,273 images across four classes, endosteal, subperiosteal, transosteal, and zygomatic, followed by hyperparameter optimization and training of the three YOLO models. Findings revealed a clear progression in model performance, with YOLOv8 achieving 93.2% accuracy and AUC-ROC of 0.94, YOLOv9 achieving 95.4% accuracy and AUC-ROC of 0.96, and YOLOv10 achieving 97.1% accuracy and AUC-ROC of 0.98. In conclusion, YOLOv10 outperformed its predecessors, establishing itself as a reliable solution for dental implant classification, thereby addressing the identified research gap.