Radiological age assessment based on clavicle ossification in CT: enhanced accuracy through deep learning

Area of Science:

• Radiology
• Artificial Intelligence
• Forensic Science

Background:

• Traditional radiological age assessment methods have limited accuracy due to discrete skeletal maturation stages.
• Continuous age estimation offers a more precise alternative for forensic and clinical applications.

Purpose of the Study:

• To develop and evaluate a deep learning model for continuous age assessment using clavicle ossification from computed tomography (CT) scans.
• To compare the model's performance against human reader estimates for established age assessment methods.

Main Methods:

• Retrospective collection of thoracic CT scans from individuals aged 15.0 to 30.0 years.
• Automatic cropping of medial clavicular epiphyseal cartilages from CT scans.
• Training a deep learning model to predict chronological age and evaluating performance using Mean Absolute Error (MAE).

Main Results:

• The deep learning model achieved a Mean Absolute Error (MAE) of 1.65 years on the test set.
• Model performance was comparable to human reader estimates (MAE of 1.84 years).
• Potential performance variations were noted due to anatomical norm-variants or pathologies.

Conclusions:

• A deep learning approach enables continuous age prediction from CT scans of the medial clavicular epiphyseal cartilage.
• This AI-driven method demonstrates performance comparable to human expert readers.
• The study highlights the potential of deep learning to enhance the accuracy of radiological age assessment.