Adults Ischium Age Estimation Based on Deep Learning and 3D CT Reconstruction

Area of Science:

• Forensic Anthropology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate age estimation is crucial in forensic and anthropological contexts.
• Traditional methods for skeletal age estimation can be invasive and time-consuming.
• Automated methods using medical imaging offer a potential solution for efficient age assessment.

Purpose of the Study:

• To develop and evaluate a deep learning model for automated age estimation.
• The model is based on 3D CT reconstructed images of the ischial tuberosity.
• The study focuses on the Han population in western China.

Main Methods:

• A retrospective dataset of 1,200 pelvic CT scans (20.0-80.0 years) was used.
• 3D virtual bone models were created, and ischial tuberosity images were extracted.
• The ResNet34 model with transfer learning was employed for age prediction.

Main Results:

• Bilateral models outperformed unilateral ones, and transfer learning improved accuracy.
• For bilateral transfer learning models: Male MAE was 7.74 years, RMSE 9.73 years; Female MAE 6.27 years, RMSE 7.82 years.
• Mixed-sex models achieved an MAE of 6.64 years and RMSE of 8.43 years.

Conclusions:

• A deep learning model using ischial tuberosity images can effectively estimate adult skeletal age.
• The ResNet34 model combined with transfer learning demonstrates feasibility and reliability.
• This automated approach offers a valuable tool for age estimation in the Han population.