An augmentation-enhanced transformer model for forensic identification using panoramic dental x-rays | JoVE Visualize

Area of Science:

Forensic Science
Biometrics
Artificial Intelligence

Background:

Accurate human identification is crucial for forensic investigations.
Traditional biometric methods and existing deep learning approaches struggle with limited data and post-mortem changes in forensic dental identification.
Panoramic dental radiographs are essential for identification when other methods fail.

Purpose of the Study:

To develop an advanced anatomical identification system, DentID-Net, for both living and deceased individuals using dental records.
To address the limitations of current deep learning methods in forensic dental identification, particularly concerning data scarcity and post-mortem alterations.
To create a robust and efficient system for automatic human identification under real-world forensic conditions.

Main Methods:

DentID-Net integrates three key components: DentAug-Min for data enhancement and overfitting reduction, a dual-branch encoder (CNNs and Transformers) for modeling dental structures and jaw shape, and IDMatch-Caps for identity matching using morphological consistency loss.
The system utilizes morphology-based enhancement to handle limited data scenarios and reduce overfitting risk.
A dual-branch encoder captures both fine dental details and overall jaw morphology, while a capsule-based matcher ensures anatomical consistency.

Main Results:

DentID-Net achieved a 98.9% mean accuracy, 99.2% specificity, and 0.999 AUC via five-fold cross-validation.
The system demonstrated superior performance compared to recent forensic dental identification systems.
Results indicate DentID-Net's ability to generalize to new data, maintain low false match rates, and operate efficiently in forensic settings.

Conclusions:

DentID-Net provides a robust and understandable system for automatic human identification using dental methods in forensic contexts.
The developed system overcomes limitations of existing methods, offering high accuracy and reliability for forensic dental identification.
The system is suitable for practical application in forensic laboratories, enhancing identification capabilities for both living and deceased individuals.