About
Clinical-grade bone density analysis, made accessible
Osteofy brings reliable, WHO-standard bone mineral density assessment to clinics and hospitals — reducing reporting time and improving patient outcomes.
Making bone density screening faster, more consistent, and widely available
Osteoporosis affects over 200 million people worldwide. Early detection through bone mineral density (BMD) assessment is critical — yet access to DXA analysis remains limited in many healthcare settings due to cost, equipment availability, and reporting bottlenecks.
Osteofy was built to bridge this gap. Our platform provides automated BMD measurement and WHO-classified reporting from standard medical imaging, enabling clinicians to make faster, data-driven decisions about osteoporosis screening and management.
2,540,000+
Scans Processed
<30s
Avg Processing Time
Upcoming
Healthcare Centres
99.7%
System Uptime
Automated BMD Extraction
Our system identifies regions of interest from DICOM and X-ray images, computes areal BMD (g/cm²), and derives T-Score and Z-Score against age-matched reference populations.
WHO Classification
Results are classified per WHO standards: Normal (T ≥ −1.0), Osteopenia (−2.5 < T < −1.0), and Osteoporosis (T ≤ −2.5), with colour-coded clinical reports.
Clinical Validation
Our analysis pipeline has been validated against reference DXA systems including Hologic and Lunar. All results are intended as clinical decision-support tools.
Methodological Approach
Chest X-ray (CXR) analysis for bone mineral density assessment represents an opportunistic screening methodology that leverages advanced deep learning algorithms to extract osseous information from standard posteroanterior (PA) and lateral chest radiographs. This approach utilizes imaging already obtained for other clinical indications, requiring no additional radiation exposure, patient time, or specialized equipment beyond standard radiological infrastructure.
The thoracic vertebrae (T4–T12) visible in standard chest X-rays provide sufficient trabecular bone information for BMD quantification. These vertebrae contain predominantly trabecular bone (60–80% by volume), which demonstrates earlier and more pronounced changes in bone mineral content compared to cortical bone. Trabecular bone has a higher metabolic turnover rate (8–10% per year vs. 2–3% for cortical), making it a sensitive indicator of systemic bone loss and metabolic bone disease.
The radiographic appearance of bone density correlates with actual mineral content through the relationship between X-ray attenuation and calcium hydroxyapatite concentration. Our algorithms analyze these radiographic density patterns to infer quantitative BMD values, calibrated against reference dual-energy X-ray absorptiometry (DXA) measurements.
Technical Implementation
Our system employs anatomy-aware multi-region-of-interest (multi-ROI) deep learning models that integrate both local anatomical features and global image context. The analysis pipeline operates through the following stages:
1. Image Preprocessing & Quality Control
DICOM images undergo automatic preprocessing including intensity normalization, contrast enhancement, and artifact detection. Quality metrics assess image sharpness, exposure adequacy, and anatomical coverage. Images failing quality thresholds are flagged for manual review or excluded from automated analysis.
2. Automatic ROI Detection & Segmentation
Deep convolutional neural networks (CNNs) with U-Net architecture identify and segment vertebral bodies and associated osseous structures. The models utilize:
- Multi-scale feature extraction to handle anatomical variations across patient populations
- Attention mechanisms to focus on vertebral bodies while suppressing overlying structures (ribs, cardiac silhouette)
- Robust segmentation that accounts for scoliosis, rotation, and partial vertebral visibility
3. Feature Extraction & Analysis
Transformer-based encoders process both local bone texture patterns and global radiographic characteristics:
- Trabecular architecture: Analysis of trabecular spacing, thickness, and connectivity patterns that correlate with bone strength
- Radiographic density gradients: Quantification of pixel intensity distributions within vertebral bodies, normalized for exposure parameters
- Bone-to-soft-tissue contrast ratios: Measurement of relative attenuation differences that reflect mineral content
- Morphological features: Vertebral height, shape, and endplate characteristics that may indicate compression fractures
4. BMD Quantification & Calibration
Ensemble learning methods combine image-derived features with patient demographic data (age, sex, ethnicity) to predict areal bone mineral density (BMD) in g/cm²:
- Multiple model architectures (CNNs, transformers, gradient boosting) are ensembled to improve robustness
- Calibration against reference DXA measurements using large-scale training datasets (10,000+ paired CXR-DXA studies)
- Population-specific adjustments for Asian, Caucasian, and other ethnic groups based on established reference databases
- Uncertainty quantification to provide confidence intervals for BMD predictions
5. Score Derivation & Classification
Statistical normalization produces clinically interpretable scores:
- T-Score: Calculated as (Patient BMD − Peak Bone Mass) / Standard Deviation, where peak bone mass is population-specific (typically age 30). Standard deviation normalization uses reference population SD (~0.12 g/cm² for Asian populations).
- Z-Score: Derived as (Patient BMD − Age-Matched Mean BMD) / Standard Deviation, enabling comparison to age-matched peers and identification of secondary causes of bone loss (Z-score < −2.0 suggests secondary osteoporosis).
- WHO Classification: Automatic categorization per WHO criteria: Normal (T ≥ −1.0), Osteopenia (−2.5 < T < −1.0), and Osteoporosis (T ≤ −2.5).
Validation & Performance Metrics
The methodology has been validated against dual-energy X-ray absorptiometry (DXA), the current gold standard for BMD measurement, through multicenter validation studies:
Reproducibility: The system ensures consistency through deterministic algorithms and result caching. Identical images with the same patient demographics produce identical BMD, T-score, and Z-score values, enabling reliable longitudinal monitoring and reducing inter-reader variability.
Clinical Considerations & Limitations
While chest X-ray-based BMD analysis provides valuable screening information, several factors influence measurement accuracy and clinical interpretation:
Image Quality Requirements
- Standard radiographic technique with appropriate kVp (70–120) and mAs settings
- Proper patient positioning (upright, full inspiration) to minimize rotation and maximize vertebral visibility
- Minimal motion artifact and adequate exposure to visualize trabecular detail
- DICOM format with preserved pixel intensity values for accurate density analysis
Anatomical Factors
- Adequate visualization of thoracic vertebrae (T4–T12) is essential for reliable analysis
- Severe scoliosis (>30° Cobb angle) may limit accuracy due to vertebral rotation
- Prior spinal instrumentation, compression fractures, or overlying structures may affect measurements
- Incomplete vertebral visualization or significant rotation reduces analysis confidence
Population-Specific Calibration
- Reference databases are calibrated for specific ethnic populations (Asian, Caucasian, Hispanic, African-American)
- Bone mineral density varies by ethnicity: Asian populations typically show 5–10% lower BMD than Caucasian reference values
- Peak bone mass and age-related decline rates differ between populations
- Results should be interpreted using population-appropriate reference standards
Clinical Interpretation
- Results should be interpreted in conjunction with clinical risk factors (fracture history, medications, comorbidities)
- When clinical suspicion is high or results are equivocal, confirmatory DXA scanning remains recommended
- Longitudinal monitoring requires consistent imaging parameters and patient positioning
- Results are intended as clinical decision-support tools and should be reviewed by qualified physicians
Clinical Note: Chest X-ray BMD analysis serves as a screening and clinical decision-support tool. It does not replace DXA for definitive diagnosis but enables opportunistic screening in patients already undergoing chest radiography, potentially identifying individuals who would benefit from formal DXA evaluation. This approach is particularly valuable in resource-limited settings or when DXA access is constrained, enabling broader population screening for osteoporosis risk.
HIPAA Compliant
All data transmission uses TLS 1.3 encryption. Patient data is encrypted at rest with AES-256. We sign Business Associate Agreements with enterprise customers.
SOC 2 Type II
Our infrastructure undergoes regular third-party security audits. We maintain SOC 2 Type II compliance for data security, availability, and confidentiality.
GDPR Ready
Data processing meets GDPR requirements for EU-based customers. Data residency options available for organisations with geographic data requirements.
Access Controls
Role-based access control (RBAC) for multi-user environments. Audit logging tracks all data access and modifications for compliance reporting.