Automated MRI-based framework combining deep learning localization and machine learning classification for pubertal...

Area of Science:

• Radiology
• Artificial Intelligence
• Pediatric Endocrinology

Background:

• Accurate assessment of skeletal maturity is crucial for clinical decisions.
• Current methods for bone age (BA) staging can be subjective and time-consuming.
• Olecranon ossification is a key indicator of pubertal development.

Purpose of the Study:

• To develop and validate a magnetic resonance imaging (MRI)-based artificial intelligence (AI) framework for automated pubertal olecranon bone age (BA) staging.
• To evaluate the internal performance of the AI framework.
• To assess the preliminary cross-centre transportability of the developed AI model.

Main Methods:

• Retrospective study utilizing elbow MRI data from two hospitals (development and external validation cohorts).
• Development of an AI pipeline involving YOLO-v8m for olecranon ROI localization and ResNet-50 with an NN classifier for BA staging.
• The pipeline was applied to the external cohort without retraining or modification.

Main Results:

• Automated olecranon localization achieved high precision (mean average precision of 1.0).
• The AI framework demonstrated strong internal validation accuracies (94.12% for girls, 93.75% for boys).
• External validation showed promising results with accuracies of 86.67% for girls and 90.48% for boys, along with high agreement metrics (kappa, ICC).

Conclusions:

• The MRI-based AI framework offers stage-consistent performance for internal validation.
• Preliminary evidence suggests good cross-centre transportability for pubertal olecranon BA staging.
• The tool serves as a supplementary maturity marker, not a standalone forensic age estimation tool.