YOLO11m-cls applied to sex and age classification based on the radiographic analysis of the nasal aperture.

Deep learning tools based on computer vision have emerged as alternative methods for assessing radiographic image patterns. These approaches have been explored for various forensic applications, including sex and age estimation. This study aimed to evaluate the diagnostic accuracy of a Convolutional Neural Network (CNN) in classifying radiographic images by sex and age, focusing on the nasal aperture as the morphological feature of interest. The sample comprised 9,349 radiographs annotated for the nasal aperture region. A CNN architecture based on the You Only Look Once series-specifically the intermediate version 11 for object classification (YOLO11m-cls)-was implemented, with training performed using 5-fold cross-validation. The overall accuracy rate was 74% (ranging from 61% to 88%), and the area under the Receiver Operating Characteristic (ROC) curve was 0.74. Correct classification rates were 73% for males and 75.17% for females. Accuracy varied with age, showing a 10% decrease among younger individuals compared to older ones. The study confirmed the reduced expression of sexually dimorphic traits in younger individuals and supported existing recommendations against performing sex estimation in subadults. Within the present methodological framework, the nasal aperture demonstrated limited applicability for sex estimation in the studied sample, with an accuracy rate corresponding to approximately one misclassification out of every four predictions.