The Sensitivity and Specificity of YOLO V4 for Tooth Detection on Panoramic Radiographs

Eha Renwi Astuti, Ramadhan Hardani Putra, Dina Karimah Putri, Nastiti Faradilla Ramadhani, Tengku Natasha Eleena Binti Tengku Ahmad Noor, Bintang Rahardjo Putra, Adhela Maheswari Pikantara Djajadiningrat

Research output: Contribution to journalArticlepeer-review


This study aimed to evaluate the performance of You Only Look Once (YOLO) v4 architecture for tooth detection on panoramic radiographs by calculating the sensitivity and specificity of a trained model. This observational descriptive study included 400 and 100 panoramic radiograph datasets that were divided into training and test data, respectively. Thirty-two permanent tooth objects were annotated based on the Fédération Dentaire Internationale numbering system. The annotated images were fed into a YOLO v4 model for the training process. Then, the trained model was tested on 100 panoramic images, which had 1,600 teeth and 1,600 edentulous areas. The sensitivity and specificity of YOLO v4 were calculated using a confusion matrix validated manually by a dental radiologist. YOLO v4 produced 1.534 and 1.568 true positive and true negative detections, respectively. The sensitivity and specificity of YOLO v4 for tooth detection on the panoramic radiographs were 99.42% and 87.06%, respectively. Within the limitations of this study, YOLO v4 demonstrated high sensitivity for tooth detection on panoramic radiographs. Further improvement in specificity should focus on minimizing the number of false positives in tooth detection through dataset improvement and architecture modification.

Original languageEnglish
Pages (from-to)442-446
Number of pages5
JournalJournal of International Dental and Medical Research
Issue number1
Publication statusPublished - Jan 2023
Externally publishedYes


  • Tooth detection
  • YOLO v4
  • artificial intelligence
  • medicine
  • panoramic radiographs


Dive into the research topics of 'The Sensitivity and Specificity of YOLO V4 for Tooth Detection on Panoramic Radiographs'. Together they form a unique fingerprint.

Cite this