Detection and Segmentation of Dental Features from Tomographic Images for Age Estimation

• Alexandre Vieira Pereira Pacelli UFJF
• Davi Magalhães Pereira UFJF
• William José Lopes Junior UFJF
• Saulo Moraes Villela UFJF
• Heder Soares Bernardino UFJF
• Karina Lopes Devito UFJF
• Marcelo Bernardes Vieira UFJF

Resumo

The age estimation represents a critical component in the forensic identification of individuals and decomposed remains. Dental analysis offers exceptional potential for age determination because teeth are the most durable structures in the human body and exhibit predictable age-related morphological changes. This study presents a comprehensive computer vision approach that uses deep learning models to detect and segment teeth in Cone Beam Computed Tomography images, extracting geometric and deep feature representations for accurate age prediction. Our approach is evaluated on a challenging dataset composed exclusively of adult subjects (ages 18-60), which are typically harder to estimate. Despite this, the method achieves low Mean Absolute Error values, demonstrating competitive performance compared to the existing literature.

Palavras-chave: Graphics, Semantic segmentation, Forensics, Pipelines, Neural networks, Estimation, Teeth, Predictive models, Feature extraction, Dentistry