Deep Learning-Based Detection of Separated Root Canal Instruments in Panoramic Radiographs Using a U2-Net Architecture

İnönü N., Aksoy U., Kırmızı D., Aksoy S., Akkaya N., ORHAN K.

Diagnostics, cilt.15, sa.14, 2025 (SCI-Expanded) identifier identifier identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 15 Sayı: 14
  • Basım Tarihi: 2025
  • Doi Numarası: 10.3390/diagnostics15141744
  • Dergi Adı: Diagnostics
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, EMBASE, INSPEC, Directory of Open Access Journals
  • Anahtar Kelimeler: artificial intelligence, deep learning, panoramic radiographs, separated endodontic instruments, U2-Net
  • Ankara Üniversitesi Adresli: Evet

Özet

Background: Separated endodontic instruments are a significant complication in root canal treatment, affecting disinfection and long-term prognosis. Their detection on panoramic radiographs is challenging, particularly in complex anatomy or for less experienced clinicians. Objectives: This study aimed to develop and evaluate a deep learning model using the U2-Net architecture for automated detection and segmentation of separated instruments in panoramic radiographs from multiple imaging systems. Methods: A total of 36,800 panoramic radiographs were retrospectively reviewed, and 191 met strict inclusion criteria. Separated instruments were manually segmented using the Computer Vision Annotation Tool. The U2-Net model was trained and evaluated using standard performance metrics: Dice coefficient, IoU, precision, recall, and F1 score. Results: The model achieved a Dice coefficient of 0.849 (95% CI: 0.840–0.857) and IoU of 0.790 (95% CI: 0.781–0.799). Precision was 0.877 (95% CI: 0.869–0.884), recall was 0.847 (95% CI: 0.839–0.855), and the F1-score was 0.861 (95% CI: 0.853–0.869). Conclusions: These results demonstrate a strong overlap between predictions and ground truth, indicating high segmentation accuracy. The U2-Net model showed robust performance across radiographs from various systems, suggesting its clinical utility in aiding detection and treatment planning. Further multicenter studies are recommended to confirm generalizability.