Clinical Scorecard: Evaluation of Automated Bone Age Determination in a German Pediatric Population: Comparison of AI Software with the Greulich and Pyle Manual Method

At a Glance

Key Highlights

• Bone age is a key index for skeletal maturity assessment, influenced by sex, nutrition, genetics, and diseases.
• Manual Greulich and Pyle method is widely used but subject to intra- and inter-observer variability.
• The CE-certified AI software IB Lab PANDA automates bone age assessment based on the G&P method using hand radiographs.

Guideline-Based Recommendations

Diagnosis

• Use left hand and wrist posterior-anterior radiographs for bone age assessment.
• Apply Greulich and Pyle atlas as the reference standard for manual bone age determination.
• Consider automated AI software as an adjunct tool for bone age estimation within validated age ranges.

Management

• Integrate AI-based bone age assessment to potentially improve consistency and reduce evaluation time.
• Use expert radiologist review to confirm AI results, especially in cases of discrepancy or poor image quality.

Monitoring & Follow-up

• Perform repeated assessments by multiple expert readers to establish ground truth in clinical validation.
• Monitor AI software performance using statistical measures such as mean absolute difference, root mean square error, and intraclass correlation coefficient.

Risks

• Potential variability in bone age estimation due to image quality or positioning errors.
• Limitations of AI software applicability outside the intended age range or populations not represented in training data.

Patient & Prescribing Data

Children aged 1 to 18 years; AI software intended for girls 36-192 months and boys 36-204 months

AI software trained on over 12,000 anonymized US pediatric hand radiographs interpreted by expert radiologists; provides bone age estimation to nearest month with visual region of analysis.

Clinical Best Practices

• Ensure high-quality hand and wrist radiographs with proper positioning for accurate bone age assessment.
• Use multiple expert readers to establish consensus bone age for validation studies.
• Employ AI software as a supplementary tool to improve objectivity and reduce intra- and inter-observer variability.
• Validate AI software performance in local populations before clinical implementation.

References

• Greulich and Pyle Atlas
• Martin et al. on Bone Age Assessment Preferences
• IB Lab PANDA CE Certification and AI Software Description