Dimorfismo sexual em exames de imagem de estruturas neurocranianas: revisões sistemáticas, análise morfológica e morfométrica tridimensional da sela túrcica e forame magno e predição do sexo por aprendizado de máquina

Abstract

Sexual dimorphism in neurocranial structures plays a relevant role in estimating biological sex and is of great interest in forensic anthropology and human identification. This study aimed to evaluate sexual dimorphism based on linear, volumetric, and morphological parameters of the sella turcica (ST) and foramen magnum (FM) in multislice computed tomography (MSCT) scans, integrating systematic reviews, three-dimensional analyses, and machine learning algorithms. Initially, two systematic reviews were conducted and registered in the PROSPERO database. These reviews synthesized evidence on sexual dimorphism in neurocranial structures and on the role of the ST in sex estimation among non-syndromic individuals. Searches across multiple databases resulted in meta-analyses of linear and volumetric measurements. The findings revealed significantly smaller neurocranial dimensions in females, with high reliability for measurements obtained through tomography. For the ST, length and diameter differed between sexes, although the certainty of the evidence was rated very low, highlighting the need for further studies with greater methodological standardization. In the analytical phase, 200 MSCT scans from adult individuals from northern and northeastern Brazil were examined to assess ST morphology according to the classifications proposed by Axelsson and Yasa. The normal, oblique anterior, round, and flat shapes were associated with sex among individuals under 30 years old, underscoring the potential of ST morphology as a complementary marker of sexual dimorphism. Subsequently, three-dimensional measurements of the ST and FM were obtained from 228 adult scans using semi-automatic segmentation and standardized cranial orientation with ITK-SNAP and 3D Slicer software. Intra-examiner reproducibility was high (ICC 0.81–0.98). All FM dimensions and most ST measurements were smaller in females, with statistically significant differences. Discriminant formulas yielded an area under the curve (AUC) of 0.798 (95% CI: 0.740–0.855), and machine learning algorithms—particularly the Support Vector Machine (SVM)—achieved moderate performance, with AUC values up to 0.8351 (± 0.0888) when both structures were evaluated together. The findings confirmed the sexual dimorphism of the ST and FM, reinforcing their value as useful anatomical markers for sex estimation. The integration of three-dimensional analyses and machine learning models demonstrated potential to enhance the accuracy of sex classification and broaden the applicability of imaging techniques in forensic anthropology. Although the results are promising, consolidation of these approaches requires external validation across diverse populations and standardization of acquisition and segmentation protocols.