Validation of AI Tool for Fracture Detection in Pediatric Osteogenesis Imperfecta

Overview

This study evaluated a commercially available AI tool's ability to detect fractures in children with osteogenesis imperfecta (OI). The AI-assisted readings by radiologists improved fracture detection accuracy compared to unaided readings, highlighting AI's potential to support diagnosis in this rare pediatric condition.

Background

Osteogenesis imperfecta is a genetic disorder characterized by fragile bones and frequent fractures, affecting about 1 in 10,000 individuals. Diagnosing new fractures in OI patients is challenging due to abnormal bone appearance from repeated injury and healing. AI tools have shown promise in fracture detection but have not been extensively validated in pediatric or rare disease populations. This study aimed to assess the performance of a commercial AI fracture detection tool in children with OI without prior specific training on this population.

Data Highlights

Key Findings

• The AI tool alone demonstrated high sensitivity (85%) and reasonable specificity (78%) for fracture detection in pediatric OI patients.
• Radiologists’ sensitivity improved significantly from 72% without AI assistance to 88% with AI support.
• Specificity of radiologists slightly decreased from 85% to 83% when assisted by AI, indicating a trade-off favoring sensitivity.
• Overall diagnostic accuracy increased from 79% to 86% with AI assistance.
• The AI tool was not trained specifically on OI cases but still improved fracture detection, suggesting generalizability.
• Radiologists found AI bounding boxes helpful for identifying subtle fractures that might otherwise be missed.

Clinical Implications

Incorporating AI tools into radiologic assessment of pediatric patients with osteogenesis imperfecta can enhance fracture detection sensitivity, potentially reducing missed diagnoses. Clinicians should be aware of a slight decrease in specificity but may consider AI assistance valuable for improving diagnostic confidence in this challenging population. Further adaptation and training of AI models on rare disease datasets could optimize performance.

Conclusion

This study demonstrates that a commercially available AI fracture detection tool can improve radiologists’ diagnostic performance in children with osteogenesis imperfecta, despite not being specifically trained on this population. AI assistance may serve as a useful adjunct in clinical practice to enhance fracture detection in rare pediatric bone fragility disorders.

References

1. Osteogenesis Imperfecta Overview and Epidemiology
2. AI in Fracture Detection Meta-Analyses
3. Milvue Suite-SmartUrgences AI Tool Development
4. CLAIM Guidelines for AI in Medical Imaging