External validation of an artificial intelligence tool for fracture detection in children with osteogenesis imperfecta: a multireader study

Category	Detail
Condition	Osteogenesis imperfecta (OI), a genetic skeletal fragility disorder with increased fracture risk
Key Mechanisms	Skeletal fragility leading to multiple fractures, with bone abnormalities complicating fracture detection
Target Population	Children under 18 years with genetically confirmed osteogenesis imperfecta
Care Setting	Pediatric radiology departments in tertiary care hospitals

OI causes frequent fractures in children, often difficult to detect due to bone remodeling and healing changes.
AI tools for fracture detection have shown improved diagnostic performance but lack validation in pediatric rare diseases like OI.
This study evaluated a commercial AI fracture detection tool's performance in pediatric OI patients without prior specific training on this population.

Use consensus expert radiologist opinion as ground truth for fracture identification in OI patients.
Label fractures as acute or healing based on radiographic features such as callus or periosteal reaction.
Consider AI assistance as an adjunct to radiologist interpretation to potentially improve fracture detection.

Integrate AI tools cautiously in clinical practice for pediatric OI fracture detection, acknowledging current lack of specific training on this population.
Maintain expert radiologist oversight when interpreting AI outputs, especially in rare disease contexts.

Conduct multireader studies with washout periods to assess AI impact on diagnostic accuracy.
Monitor AI performance continuously to identify potential biases or limitations in pediatric rare disease fracture detection.

Potential for missed fractures due to bone remodeling and atypical appearances in OI.
Risk of AI misclassification or reduced sensitivity in populations not included in training datasets.
Possibility of patient discrimination or bias if AI tools are implemented without validation in rare disease groups.

Pediatric patients with genetically confirmed osteogenesis imperfecta undergoing radiographic fracture evaluation

AI tools may assist radiologists in detecting fractures but require validation and cautious interpretation in this population.

Use a multidisciplinary approach combining expert radiologist consensus and AI assistance for fracture detection in OI.
Provide radiologists with training on AI tool use and fracture definitions prior to implementation.
Ensure anonymization and secure handling of patient imaging data during AI analysis.
Implement washout periods in studies to reduce recall bias when assessing AI impact.
Interpret AI outputs as supportive rather than definitive, especially when AI cannot classify fracture healing status.

Clinical Scorecard: Validation of an AI-Based Tool for Detecting Fractures in Pediatric Patients with Osteogenesis Imperfecta: A Multireader Analysis