Automatic identification of individuals using deep learning method on panoramic radiographs

Area of Science:

• Forensic Dentistry
• Artificial Intelligence
• Radiology

Background:

• Dental structures possess unique individual characteristics and remain stable postmortem, making them valuable for forensic identification.
• Traditional forensic dental identification methods can be time-consuming and require specialized expertise.

Purpose of the Study:

• To develop an automated individual identification system utilizing deep learning algorithms applied to panoramic radiographs (PRs).
• To assess the efficacy and precision of various deep learning models for dental identification.

Main Methods:

• A dataset of 4966 panoramic radiographs (PRs) from 1663 individuals was compiled, with 3303 images used for training.
• Five deep learning models (Vgg16, Vgg19, ResNet50, ResNet101, EfficientNet) were implemented and evaluated for their identification performance.

Main Results:

• The Vgg16 model demonstrated the highest precision, achieving approximately 80-90% accuracy over 200 epochs.
• Identification was successful using the Top-N metrics (5-15 candidate labels) within 5-10 seconds, even with limited dental features.
• All tested models showed varying degrees of precision in matching individuals based on their dental radiographs.

Conclusions:

• The developed deep learning-based identification system using PRs is a promising and potentially useful tool for forensic applications.
• This system can aid in the identification of unidentified bodies and missing persons, including elderly individuals.
• The technology offers significant benefits for law enforcement, government agencies, and disaster response efforts.