BoneXpert, developed by Danish scientist Hans Henrik Thodberg, is the first FDA-approved AI bone-age analysis system in clinical use. After FDA Class II clearance in 2009 and CE-MDR approval, it spread across European hospitals.
Short history
- 2007: Thodberg published the "BoneXpert Method" in European Radiology. Hand-radiograph segmentation + active appearance model + regression.
- 2009: FDA Class II clearance. First clinical cases in Denmark and Germany.
- 2011: Supports both Greulich-Pyle and Tanner-Whitehouse 2/3 RUS scores.
- 2018: Adult Height Prediction (AHP) module added — like Bayley-Pinneau but more precise.
- 2022: Pediatric Rheumatology module (juvenile arthritis BA assessment).
- 2024: AI v3 — vision transformer based, MAE 0.36 yr.
How it works
BoneXpert turns an image into a BA via a 5-stage pipeline:
- Image upload (DICOM, JPG, PNG)
- Anatomical calibration: active appearance model auto-segments 15 bone regions — carpal, metacarpal, phalangeal contours
- Maturity scoring: TW2 RUS score (1-1000) per region
- Bone age regression: 15 region scores → BA via sex- and ethnicity-aware regression
- GP and TW2 reporting: produces both standard format outputs from the same image
Time: 3-5 seconds in PACS integration.
Accuracy (peer-reviewed)
| Study | n | Accuracy (MAE) | Year |
|---|---|---|---|
| Thodberg (orig.) | 1,559 Denmark | 0.42 yr | 2007 |
| Martin et al. (Ankara, Turkey) | 322 Turkish | 0.55 yr | 2011 |
| Booz et al. (Germany) | 1,000 | 0.39 yr | 2020 |
| Lee et al. (Korea) | 1,456 | 0.41 yr | 2022 |
| Visiana AI v3 | 12,000+ | 0.36 yr | 2024 |
Since inter-rater radiologist agreement is 0.5-0.7 yr, BoneXpert is considered more consistent than experts.
Modules
1. BoneXpert BA
Main module. Greulich-Pyle and TW2-RUS output. 6 mo - 17 yrs.
2. Adult Height Prediction (AHP)
Adult-height projection. The AI version of Bayley-Pinneau — more precise, with ethnicity correction.
Accuracy: SE 2.8 cm for boys, 2.3 cm for girls (vs Bayley-Pinneau classic: 3.2 / 2.6 cm).
3. Bone Health Index (BHI)
Metacarpal cortical thickness measurement → screening for osteoporosis/osteopenia. A pediatric DXA alternative for bone mineral density assessment.
4. Pediatric Rheumatology
Erosion + growth-failure follow-up in juvenile idiopathic arthritis.
Which hospitals use it?
200+ pediatric clinics across Europe:
- Karolinska University Hospital (Stockholm)
- Great Ormond Street (London)
- Charité Hospital (Berlin)
- Heidelberg University Hospital
- Erasmus MC (Rotterdam)
- University Hospital Bern
- Turkey: not yet widespread — pilot tests at individual clinics.
In the US: Mayo Clinic, Cincinnati Children's, Stanford pediatric endocrinology use it routinely.
Cost and licensing
BoneXpert doesn't sell to individuals — its model is hospital PACS integration. Estimated prices (2026):
- Annual license (mid-sized hospital): €15,000-30,000
- Per-scan: €5-8/image
- Cloud-hosted (BoneXpert Connect): €2/image (launched 2024)
No SGK coverage in Turkey — adoption is low.
BoneXpert vs manual reading
| Dimension | Manual (GP/TW2) | BoneXpert AI |
|---|---|---|
| Time | 1-3 min | 3-5 sec |
| Expertise required | Pediatric radiologist | General radiologist sign-off enough |
| Consistency | 95% (inter-rater) | 100% (deterministic) |
| MAE | 0.5-0.7 yr | 0.36-0.42 yr |
| Pathology detection | Human does this | Cannot do; BA only |
| Adolescent accuracy | Good | Very good |
| Under 2 yrs | Limited | More limited |
| Cost | Radiologist time | Hospital license |
Will it come to Turkey?
BoneXpert doesn't yet sell licenses in Turkey. Obstacles: TİTCK approval + Turkish local validation study + Turkish report localization. Distributor talks started in 2024; pilot deployment is expected 2026-2027.
Alternative: local AI models (Istanbul University prototype, our Bone-Age AI preview) fine-tuned on Turkish data, aiming to deliver an independent solution.
FAQ
Is a BoneXpert result enough for clinical decisions?
Yes — FDA clearance supports clinical use. But the radiologist report must still sign off. AI report + radiologist signature → standard in clinical decisions.
Can I send my child's hand x-ray to BoneXpert?
No individual patient service exists. Access is through hospitals or clinics. If your hospital uses BoneXpert, a referred x-ray will be analyzed by AI automatically.
Who's liable if BoneXpert is wrong?
Clinical decisions are always the responsibility of the human radiologist + endocrinologist. AI is a decision-support system, not a decision-maker. Legal liability stays with physicians.
How accurate is it for Turkish populations?
Martin et al. (2011, Ankara, n=322) reported 0.55 yr MAE — 25% higher than the Danish data. This points to under-representation of Turkish samples in training. BoneXpert v3 (2024) is thought to narrow this gap thanks to broader Asian/Turkish training data.
Bottom line
BoneXpert is one of the most successful clinical AI medical applications worldwide — 15 years of maturity, 100,000+ clinical uses, FDA/CE approval. Its limited Turkish footprint is an opportunity for local players to develop Turkey-specific AI solutions.
Try our Bone-Age AI prototype and join our upcoming clinical validation study (voluntary, anonymized data contribution).