Skull CT metadata for automatic bone age assessment by using three-dimensional deep learning framework

Area of Science:

• Forensic Anthropology
• Medical Imaging
• Artificial Intelligence

Background:

• Bone age assessment (BAA) is crucial in forensic science, particularly when only skeletal remains like skulls are available.
• Current methods face challenges in accuracy, especially in extreme forensic scenarios.

Purpose of the Study:

• To develop an accurate 3D deep learning (DL) framework for BAA using skull CT metadata.
• To explore novel skull markers for enhanced BAA accuracy.

Main Methods:

• Retrospective analysis of 385,175 skull CT slices from 1,085 patients (ages 16.32-90.56).
• Development of a 3D DL framework and comparison with existing DL and traditional machine learning (ML) models.
• Evaluation using Mean Absolute Error (MAE) on training, test, and external validation sets.

Main Results:

• The proposed 3D DL framework achieved superior MAE compared to other models: 5.70 years (males) and 7.84 years (females) on the test set.
• Traditional ML and other DL methods showed higher MAE, ranging from 10.12 to 14.12 years.
• The model identified new skeletal markers, enhancing BAA capabilities.

Conclusions:

• The developed 3D DL framework offers a more accurate and robust approach to BAA from skull CT data.
• This framework can serve as a foundation for advanced feature extraction from 3D skull CT metadata in forensic and clinical settings.
• The study highlights the potential of AI in overcoming limitations in forensic age estimation.