Performance of an artificial intelligence system for bone age assessment in Tibet

Area of Science:

• Pediatric Endocrinology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate bone age (BA) assessment is crucial for pediatric growth evaluation.
• Limited medical resources in regions like Tibet necessitate innovative assessment tools.
• Existing AI systems for BA are often developed using data from different ethnic groups and geographic locations.

Purpose of the Study:

• To evaluate the accuracy of a fully automated artificial intelligence (AI) system for bone age (BA) assessment in children residing in the Tibet Highland.
• To compare the AI system's performance against expert-determined BA using the Greulich and Pyle (GP) method.

Main Methods:

• Collected left hand radiographs from 385 children (4-18 years) in Tibet.
• Determined BA by expert consensus using the Greulich and Pyle (GP) method.
• Estimated BA using a previously validated AI system developed for Han Chinese children.

Main Results:

• The AI system achieved an accuracy of 84.67% for Tibetan children and 89.41% for Han children within a 1-year margin.
• The mean absolute difference (MAD) between AI and expert assessments was 0.65 years for Tibetan children and 0.56 years for Han children.
• Discrepancies in hand-wrist bone maturation were identified as a key factor affecting accuracy in younger children (4-6 years).

Conclusions:

• An AI BA system developed for Han Chinese children can accurately assess bone age in Tibet, despite geographical and ethnic differences.
• The AI-based BA system presents a viable, efficient, and effective solution for bone age assessment in Tibet's resource-limited settings.
• This technology holds promise for improving pediatric care in underserved regions.