Liodon Health
Clinical AI that stays
inside your clinic.
Open-source diagnostic tools built on research-grade models.
Patient data never leaves your hardware. No cloud. No subscription. No compromise.
Our approach
Healthcare AI has a data problem. We solve it by running entirely on-device.
Privacy-first
Zero data egress
All inference runs locally on your machine. No patient images, no metadata, no telemetry is ever transmitted. HIPAA and GDPR compliance is structural, not a setting.
Open weights
Fully auditable
Model weights, training datasets, and inference code are all public. You can inspect exactly what the model learned and on what data — no black-box diagnostics.
Lightweight models
Runs on clinic hardware
Designed for the CPU in the front desk computer, not a data centre. No GPU required. Models are compact enough to ship as single-file binaries alongside the application.
Products
Dental Radiology · Available now
Liodon Dental
AI-powered panoramic X-ray analysis that runs entirely on your hardware. Detects caries, periapical lesions, and impacted teeth using a YOLO11-N model with 2.6M parameters — small enough to run on a CPU, fast enough for clinical workflow.
- 63.0% mAP50 on DENTEX validation set
- Trained on 9,928 panoramic X-rays (DENTEX + OralXrays-9)
- Single binary — Windows & Linux, no install, no internet
- Model weights and dataset fully open on HuggingFace
What’s coming
Liodon Dental is the first product in the Liodon Health line. More specialties are in research.
In research
Radiology screening
Chest X-ray triage and anomaly flagging for resource-constrained clinics. Targeting common findings — effusion, consolidation, cardiomegaly — with transparent confidence scores.
In research
Pathology assistance
Slide-level classification for common histopathology findings. Designed to run on digitised slides without requiring cloud upload or proprietary scanners.
In research
Clinical note structuring
Local SLM-powered extraction of structured data from free-text clinical notes. Runs on-device using quantized models from the Liodon model library.
Open source, always.
Every model, every dataset, every line of inference code is public. Fork it, audit it, improve it.