YOLOv7 detects periapical lesions on pediatric panoramic radiographs with higher sensitivity than dental students

A YOLOv7 deep learning model detected periapical lesions on pediatric panoramic radiographs with sensitivity 76.1% vs 55.2% for dental students, specificity 99.8% vs 97.3%, in a small internal validation study.

• The model was trained on 302 radiographs (227 original plus 75 augmented with grayscale, noise reduction, and edge detection) and tested on a larger held-out set.
• Precision (99.8% vs 95.3%) and F1 score (86.4% vs 69.9%) also favored the algorithm.
• Limitations include a small sample size, retrospective design, and internal validation only; no external test set was evaluated.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.