An artificial intelligence-based bone age assessment model for Han and Tibetan children.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pediatric Endocrinology

Background:

• Manual bone age assessment (BAA) is time-consuming, costly, and variable.
• These challenges are amplified in resource-limited settings like the Tibetan Plateau.
• Artificial intelligence (AI) offers a potential solution for automated BAA.

Purpose of the Study:

• To develop and evaluate an AI-based BAA model for Han and Tibetan children.
• To compare the performance of the developed AI model (EVG-BANet) against a state-of-the-art model (BoNet).
• To assess the model's accuracy and generalizability across diverse datasets.

Main Methods:

• Trained the EVG-BANet model using three datasets: RSNA, RHPE, and a self-established local dataset.
• Included Han and Tibetan children in the local and external test sets.
• Evaluated models using Mean Absolute Difference (MAD) and accuracy within 1 year, comparing EVG-BANet against BoNet.

Main Results:

• EVG-BANet demonstrated superior performance over BoNet in MAD on RHPE, local, and external test sets.
• EVG-BANet achieved higher accuracy within 1 year on local (97.7%) and external (89.5%) test sets.
• The model showed no bias in the local test set but exhibited age-related bias in the external test set.

Conclusions:

• EVG-BANet accurately predicts bone age (BA) in both Han and Tibetan children.
• The AI model is suitable for use in areas with limited healthcare facilities, such as the Tibetan Plateau.
• Further refinement may be needed to address age-related biases in specific populations.