Ethmoid sinus CBCT imaging as a biometric instrument: dataset creation for deep learning identification.

Area of Science:

• Forensic Radiology
• Medical Imaging
• Deep Learning

Background:

• Cranial structures like ethmoid bones offer biometric data for human identification.
• Cone Beam Computed Tomography (CBCT) provides non-invasive visualization of these structures.

Purpose of the Study:

• To create an annotated CBCT dataset of the ethmoid bone.
• To assess its effectiveness for deep learning-based gender classification.

Main Methods:

• Collected 565 CBCT scans (312 male, 253 female, ages 6-74).
• Expert radiologists annotated ethmoid regions on axial slices.
• Trained a Convolutional Neural Network (CNN) model for gender classification.

Main Results:

• A fine-tuned ResNet-50 model achieved an 87% F1-score.
• Demonstrated strong potential for ethmoid-based gender classification.

Conclusions:

• The developed dataset is a reproducible resource for forensic radiology and AI research.
• Ethmoid CBCT imaging shows potential as a biometric marker for identification tasks.