YOLO11-N · 2.6M Parameters · CPU Only
Liodon Dental
AI-powered dental panoramic X-ray analysis, running entirely on your hardware.
Detects caries, periapical lesions, and impacted teeth. Patient data never leaves your clinic.
Windows 10+ · Ubuntu 20.04+ · Free & open source · Decision-support only — not FDA cleared
What it detects
Trained on 9,928 panoramic X-rays from DENTEX and OralXrays-9 (CVPR 2025).
Class 0
Caries
Dental decay, including deep caries. Directionally accurate — panoramic resolution is the detection ceiling, not the model.
Class 1
Periapical Lesion
Infection at root tip (apical periodontitis). Fires correctly when present; performance limited by small validation set size.
Class 2
Impacted Tooth
Impacted and wisdom teeth. Highest-quality detection among the three classes — near clinical-grade performance.
Model Performance
Validated on DENTEX val set. Best checkpoint: epoch 33 of 57 (early stopping).
63.0%
mAP50
38.8%
mAP50-95
64.4%
Precision
69.4%
Recall
How it works
Step 1
Download & Run
Download the binary for your platform and run it directly. No installation, no dependencies, no internet connection required.
Step 2
Open localhost:8080
The app starts a local web server at http://localhost:8080. Open it in any browser on the same machine.
Step 3
Drag & Drop X-Ray
Drag and drop a panoramic radiograph. The model runs inference locally and returns annotated findings in seconds.
System Requirements
Windows
- Windows 10 or later (64-bit)
- 4 GB RAM minimum, 8 GB recommended
- 500 MB free disk space
- No GPU required — CPU inference via ONNX Runtime
Linux
- Ubuntu 20.04+ or equivalent (64-bit)
- 4 GB RAM minimum, 8 GB recommended
- 500 MB free disk space
- Run with:
chmod +x liodon-dental && ./liodon-dental
Model Details
Architecture
- YOLO11-N (nano), 182 layers
- 2.6M parameters
- 6.4 GFLOPs compute
- 640×640 RGB input
Inference
- ONNX format (10.6 MB)
- CPU only — ONNX Runtime
- Letterbox preprocessing + per-class NMS
- 8,400 candidate boxes, 3 classes
Training
- 9,928 panoramic X-rays
- DENTEX + OralXrays-9 (CVPR 2025)
- Confidence threshold: 0.45
- IoU threshold: 0.35
Open Source
Model weights, training dataset, and inference code are all publicly available.
Clinical Disclaimer
This tool is intended as a decision-support aid only. All findings must be reviewed and confirmed by a licensed dental professional. Not cleared by the FDA for diagnostic use. Do not use as a sole basis for clinical decisions.
Ready to get started?
Free to download. Open-source model. No account required. Runs entirely offline.