Forensic dental age estimation with deep learning: a modified xception model for panoramic X-Ray images

Area of Science:

• Forensic dentistry
• Artificial intelligence in medicine
• Medical imaging analysis

Background:

• Accurate forensic age estimation is crucial for legal and ethical reasons.
• Orthopantomography (OPG) images offer valuable data for age assessment.
• Deep learning presents a promising approach for enhancing age estimation accuracy.

Purpose of the Study:

• To develop an improved deep learning method for forensic age estimation using OPG images.
• To differentiate individuals under 12 from those 12 and older.
• To identify the most effective deep learning model for this forensic application.

Main Methods:

• Collected a dataset of 1941 pediatric patients aged 5-15 years from two radiology departments.
• Evaluated various deep learning models including Xception, ResNet, and EfficientNet.
• Utilized traditional metrics (CA, SE, SP, K, AUC) and a novel Polygon Area Metric (PAM) for imbalanced datasets.

Main Results:

• The "Forensic Xception" model, based on Xception, achieved the highest performance with a PAM score of 0.8828.
• Demonstrated superior classification accuracy, sensitivity, specificity, Kappa, AUC, and F1 Score.
• The PAM metric provided a comprehensive evaluation, particularly for imbalanced forensic datasets.

Conclusions:

• Deep learning models, especially "Forensic Xception," significantly advance forensic age estimation from OPG images.
• This method shows potential for accurate age classification in legal contexts.
• Further research is recommended to explore larger datasets, refine models, and address ethical considerations.