Discriminação das idades legais de 14 e 18 anos por meio da morfometria geométrica facial

Abstract

Forensic age investigation plays a crucial role, presenting challenges when the only available evidence consists of photographs. Moreover, the age milestones of 14 and 18 years hold particular relevance in the legal context. This study aimed to assess the applicability of Geometric Morphometrics in discriminating critical ages in a Brazilian sample. This was a cross-sectional accuracy study considering two critical ages of legal interest (14 and 18 years). The dataset consisted of frontal photographs of Brazilian individuals aged between 6 and 22 years, obtained from the National Passport System of the Federal Police, with a sample of 17,000 images balanced by age and sex (n = 500 for each sex at each age). The photographs underwent automated facial analysis through an approach integrating concepts of computer vision and image processing methods within supervised machine learning procedures, utilizing the demarcation of 28 photoanthropometric landmarks. Based on the set of information related to the coordinates of these landmarks, geometric morphometric analysis of the face was applied using the Procrustes paradigm, followed by statistical tests to identify and quantify geometric differences within the sample. The results of the Generalized Procrustes Analysis allowed for the development of displacement graphs in relation to the critical ages of legal interest, enabling the assessment of shape discrepancies between a reference group and a target group. The coordinates generated by the Generalized Procrustes Analysis were subjected to Principal Component Analysis to explain the variance and covariance structure among the set of variables. The first principal component accounted for 48.86% of the total variability of the Procrustes coordinates, while the second component explained 20.43%, leading to a total variability explanation of 69.29%, a value considered satisfactory for this study. Logistic regression analysis was used to investigate the discriminative ability of geometric morphometrics concerning the ages of 14 and 18, based on the principal component scores of the Procrustes coordinates and the centroid, encompassing all generated principal components. Three models were developed: one for males, one for females, and one for the total sample, adopting a 5% significance level. For the age of 14 years, the method's accuracy was 87.7%, with higher performance in males (91.7%) compared to females (86.9%). Sensitivity was 89.6% (males: 92.6%; females: 89.7%), and specificity was 85.8% (males: 90.7%; females: 83.7%). For the age of 18 years, accuracy was 76.4%, again with superior results for males (83.7%) compared to females (80.0%). Sensitivity was 87.3% (males: 73.6%; females: 60.4%), and specificity was 65.7% (males: 87.9%; females: 88.2%). These findings suggest that facial geometric morphometrics is a skilled method for discriminate legal ages, with more effective performance in determine the age of 14, especially for male individuals. The results suggest that Geometric Morphometrics, integrated with automated face analysis, has the potential to become an auxiliary tool for forensic practice.