Sex estimation using machine learning algorithms and morphometric assessment of the sella turcica related parameters

Angela Graciela Deliga Schroder, Juliana Marodin Fauri Rotta, Rayane Délcia da Silva, Flares Baratto-Filho, Bianca Marques de Mattos de Araujo, Svenja Beisel-Memmert, Bianca Simone Zeigelboim, Karinna Veríssimo Meira Taveira, Natanael Henrique Ribeiro Mattos, Erika Calvano Küchler, Cristiano Miranda de Araujo,
Sex estimation using machine learning algorithms and morphometric assessment of the sella turcica related parameters,
Forensic Imaging,
Volume 43,
2025,
200654,
ISSN 2666-2256,
https://doi.org/10.1016/j.fri.2025.200654.
(https://www.sciencedirect.com/science/article/pii/S2666225625000326)
Abstract: Sexual dimorphism refers to morphological differences between sexes, with parameters related to the sella turcica showing significant variations between males and females. This study aimed to evaluate the role of sella turcica parameters in sex prediction using machine learning algorithms. Lateral cephalometric radiographs from 470 patients were analyzed to measure sella turcica dimensions, including length, depth, diameter, anterior and posterior height, width, calcification type, and the sella-nasion and sella-ponticulus lines. Machine learning models—including Logistic Regression, Gradient Boosting, K-Nearest Neighbors, Support Vector Machine (SVM), Multilayer Perceptron, Decision Tree, AdaBoost, and Random Forest—were trained and validated using 5-fold cross-validation. Model performance was evaluated by area under the curve (AUC), accuracy, recall, precision, F1 score, and ROC curves. The linear distance between the sella and ponticulus was the most significant predictor across all models. Model AUC ranged from 0.85 (95 % CI: 0.78–0.90) in testing, with similar values in cross-validation, while precision reached 0.77 in testing and 0.78 in cross-validation. The SVM model achieved a balanced performance across all metrics. In conclusion, sella turcica parameters demonstrated sexual dimorphism when used as landmarks in two-dimensional lateral images. The predictive model showed strong capability, highlighting its potential as an auxiliary tool for forensic sex identification.
Keywords: Sella Turcica; Sex characteristics; Artificial intelligence; Machine Learning