YOLO11s - Dental Caries Detection on Panoramic Radiographs
Fine-tuned YOLO11s for detecting caries lesions in dental panoramic X-rays.
Results (test set)
| metric | value |
|---|---|
| mAP50 | 0.8624 |
| mAP50-95 | 0.6333 |
| precision | 0.8378 |
| recall | 0.7986 |
| F1 | 0.8177 |
Plots are in test/. Training curves are in training/.
Data
OralXrays-9 (CVPR 2025), "Decay" category. 6,134 panoramic radiographs with 11,228 annotated lesions. Split 70/15/15 with a fixed seed (42).
Training
| parameter | value |
|---|---|
| base model | yolo11s.pt |
| epochs | 150 |
| imgsz | 1024 |
| batch | 8 |
| patience | 30 |
Usage
from huggingface_hub import hf_hub_download
from ultralytics import YOLO
p = hf_hub_download("EuricoGVP/yolo11s-caries", "model.pt")
model = YOLO(p)
results = model("radiograph.jpg")
results[0].show()
Limitations
Research model. It must NOT be used for clinical diagnosis. It has not undergone clinical validation or regulatory approval.
Trained on a single source, so it may not generalize to other X-ray machines, acquisition protocols, or patient populations.
Ground truth comes from the OralXrays-9 annotations, produced by specialists. The model reproduces those annotation criteria, which may differ from other institutions.
Citation
Original data: OralXrays-9 (CVPR 2025). Repository: https://github.com/Binz-Chan/CVPR2025_OralXrays-9
- Downloads last month
- 26