DENSEN: a convolutional neural network for estimating chronological ages from panoramic radiographs

Area of Science:

• Forensic Science
• Radiology
• Artificial Intelligence

Background:

• Age estimation from panoramic radiographs is crucial in forensic science.
• Existing methods often focus on juveniles and lack reliability for older adults.
• Previous studies relied on statistical or scoring-based approaches requiring extensive lab work.

Purpose of the Study:

• To develop a deep learning model for accurate chronological age estimation from panoramic radiographs.
• To address limitations of previous methods, particularly for adult populations.
• To create a versatile and accessible tool for age assessment.

Main Methods:

• Developed DENSEN, a deep learning model based on Soft Stagewise Regression Network (SSR-Net).
• Trained and validated the model using 1903 clinical panoramic radiographs from individuals aged 3 to 85.
• Evaluated model performance using Mean Absolute Error (MAE) across different age groups.

Main Results:

• DENSEN achieved low MAEs: 0.6885 (children), 0.7615 (teens), 1.3502 (young adults), and 2.8770 (adults).
• The model demonstrates effectiveness regardless of gender.
• DENSEN exhibits superior accuracy for adults (25+ years) compared to existing methods.
• The model is memory-efficient, requiring approximately 1.0 MB overhead.

Conclusions:

• Introduced DENSEN, a novel deep learning approach for age estimation from panoramic radiographs.
• The method requires minimal laboratory work compared to traditional techniques.
• DENSEN is an open-source tool applicable to all age groups, enhancing forensic age assessment accessibility.