Detección de lesiones periapicales en dientes con prótesis fijas mediante modelos de segmentación y análisis de relaciones corona-lesión
Video Experimental Relacionado
Updated: May 2, 2026
A Finite Element Approach for Locating the Center of Resistance of Maxillary Teeth
Published on: April 8, 2020
Buse Cebi Gul1, Can Cebi2, Ersel Malkoc3
1Assistant Professor, Department of Prosthodontics, Faculty of Dentistry, Inonu University, Malatya, Turkey.
Statement Of Problem:
Periapical lesions in teeth with fixed prostheses often remain undiagnosed in routine panoramic radiographic evaluations, leading to delayed treatment and potential tooth loss. The complex anatomy of prosthetic restorations can mask early periapical pathology, and manual detection is time-consuming and subject to interobserver variability. A standardized automated system is not currently available for the simultaneous detection of periapical lesions and the analysis of their relationship with fixed prostheses in panoramic radiographs.
Purpose:
The purpose of this study was to evaluate the diagnostic accuracy of artificial intelligence models based on the YOLO11 architecture for detecting periapical lesions in teeth with fixed prostheses on panoramic radiographs and to analyze crown-lesion relationships using automated algorithms.
Material And Methods:
A total of 1686 annotations (1033 crowns, 653 periapical lesions) were manually labeled on 404 retrospectively selected panoramic radiographs obtained from patients at Inonu University Faculty of Dentistry between March 2024 and May 2025. Manual labeling was performed independently by 2 experienced observers using the Roboflow platform. The dataset was divided into 77% training (312 images), 13% validation (52 images), and 10% testing (40 images). Five YOLO11 segmentation variants were trained for 150 epochs. Model performance was evaluated using precision, recall, mAP50, and mAP50-95 metrics. Statistical analyses were performed using Python 3.9 with scikit-learn and scipy libraries (α=.05). Receiver operating characteristic (ROC) curves and area under the curve (AUC) values were calculated for diagnostic performance assessment.
Results:
The YOLO11l-seg model achieved the highest performance with mAP50 of 0.885, recall of 0.853, and precision of 0.847. While all models demonstrated high success in crown detection (mAP50: 0.975 to 0.980), YOLO11l-seg yielded the best results for periapical lesion detection (mAP50: 0.794). Crown-lesion relationship analysis revealed that 84.62% of lesions were associated with crowns, with mandibular crowns showing a 2.7 times higher lesion prevalence than maxillary crowns (52.24% against 19.05%, P<.001).
Conclusions:
YOLO11-based artificial intelligence models demonstrated high accuracy for detecting periapical lesions in teeth with fixed prostheses. The developed Python algorithm successfully analyzed crown-lesion relationships, providing quantitative data for clinical assessment.
