AgeMiner: A novel metadata-integrated chronological age predictor based on bone mineral density data

Area of Science:

• Forensic Anthropology
• Radiology
• Biomedical Engineering

Background:

• Bone mineral density (BMD) estimation for chronological age is crucial in forensic anthropology.
• Dual-energy X-ray absorptiometry (DXA) of the distal radius and ulna is common for osteoporosis screening but underutilized in forensic science.
• Limited forensic research exists using DXA data from the distal 1/3 radius and ulna.

Purpose of the Study:

• To develop and validate machine learning models for chronological age estimation using DXA-derived BMD metrics.
• To assess the impact of metadata (sex, BMI, osteoporosis diagnoses) on age estimation accuracy.
• To integrate optimized models into an accessible tool for forensic practitioners.

Main Methods:

• Utilized a retrospective dataset of 5,134 DXA scans (ages 12-96) from the distal 1/3 radius and ulna.
• Trained various machine learning models including Linear Regression (LR), Support Vector Regression (SVR), Random Forest Regression (RFR), XGBoost (XGB), and LightGBM (LGBM).
• Optimized models using Bayesian cross-validation and evaluated performance based on Mean Absolute Error (MAE).

Main Results:

• A simple model using BMD data + Diagnoses achieved an MAE of 2.40 years.
• The best overall performance was an RFR model combining Female + Diagnoses, yielding an MAE of 2.18 years.
• An RFR model for Normal weight + Diagnoses showed an MAE of 2.54 years when considering BMI.

Conclusions:

• The AgeMiner tool integrates optimized machine learning models for efficient and accurate chronological age estimation.
• Models demonstrate customizable application based on individual metadata, enhancing forensic anthropology applications.
• The tool provides a valuable resource for age estimation in adults and the elderly using DXA BMD data.