DMF-T assessment on panoramic images using deep learning-based convolutional neural network algorithm

Automated, image-based evaluation of oral health indices can improve diagnostic consistency and support large-scale surveillance of dental disease. Leveraging recent advances in deep learning–based object detection, this study investigates a convolutional neural network approach for quantifying the Decayed, Missing, and Filled Teeth (DMFT) index from panoramic radiographs. A total of 1,316 anonymized digital panoramic images from patients aged > 18 years were retrospectively collected and annotated according to World Health Organization criteria. Using the FDI numbering system, teeth were labeled as Healthy, Decayed, or Filled, and Missing teeth were identified by comparison with a complete 32-tooth reference framework. The dataset was randomly partitioned into training, validation, and test subsets. For automated tooth-level analysis, a YOLOv11-based CNN object detection framework was implemented to simultaneously detect and classify individual teeth on panoramic images. Model performance was assessed using precision, recall, F1-score, mean average precision (mAP), and missing-tooth detection accuracy. The YOLOv11-based model achieved high overall performance for automated DMFT assessment ([email protected] = 0.839; F1-score = 0.819). Classification performance was highest for Filled teeth, whereas detection and classification of Decayed teeth were comparatively less accurate. Missing teeth were identified with an accuracy of 95.48%. These findings indicate that the proposed YOLOv11-based framework provides reliable, tooth-level DMFT estimation from panoramic radiographs and has potential as a decision-support tool for epidemiological surveillance and clinical practice. Further methodological optimization is warranted to enhance early caries detection performance.