Development of an AI-based model for sex estimation using CT-derived metrics from paranasal sinuses

Sex estimation plays a critical role in the reconstruction of the biological profile in forensic contexts. The paranasal sinuses, owing to their structural complexity and resistance to postmortem degradation, have emerged as valuable anatomical regions for this purpose. This observational, retrospective study assessed the accuracy of sex estimation in Brazilian adults using linear and volumetric measurements of the frontal, maxillary, and sphenoidal sinuses obtained from multislice computed tomography (MSCT). A total of 220 MSCT scans (113 males, 107 females) from three imaging centers were analyzed. Semiautomatic segmentation was performed using ITK-SNAP and 3D Slicer to extract craniometric features. Statistical analyses and supervised machine learning models, including logistic regression, linear support vector machine (SVM), random forest, and k-nearest neighbors, were employed for classification. Males showed significantly larger anteroposterior, supero-anterior, and volumetric dimensions (p < 0.05). The highest classification accuracy was achieved by logistic regression and linear SVM models, both reaching 84%. These models outperformed conventional discriminant analysis. The study supports the forensic utility of CT-derived morphometric parameters of the paranasal sinuses for sex estimation and highlights the potential of machine learning as a robust complementary tool in forensic investigations. Findings also underscore the relevance of broader validation in multiethnic forensic contexts.