Radiographic morphology of canines tested for sexual dimorphism via convolutional-neural-network-based artificial...

Area of Science:

• Forensic Anthropology
• Dental Forensics
• Artificial Intelligence in Medicine

Background:

• Permanent left mandibular canines are utilized for sex determination in forensic identification.
• The efficacy of canine morphology for distinguishing sex remains debated.

Purpose of the Study:

• To evaluate the sexual dimorphism of permanent left mandibular canines using an artificial intelligence (AI) approach.
• To assess the accuracy of AI in predicting sex based on canine morphology across different age groups.

Main Methods:

• A dataset of 13,046 radiographic images of left mandibular canines from 5,838 males and 7,208 females (ages 6-22.99) was analyzed.
• The DenseNet121 model was employed for binary sex classification across 17 one-year age categories.
• Performance was quantified using accuracy rates, receiver operating characteristic (ROC) curves, and confusion matrices.

Main Results:

• AI classification accuracy ranged from 57% to 76%, with a mean of 68% ±5%.
• The area under the ROC curve (AUC) varied between 0.58 and 0.77.
• Optimal performance was observed around age 12, with lower accuracy near age 7.

Conclusions:

• The AI-based analysis indicates moderate, but not definitive, sexual dimorphism in left mandibular canines.
• Canine morphology analysis for sex estimation should be a secondary method, used only when other dimorphic features are unavailable.