Artificial intelligence model for predicting sexual dimorphism through the hyoid bone in adult patients.

Abstract

The objective of this study was to develop a predictive model using supervised machine learning to determine sex based on the dimensions of the hyoid bone. Lateral cephalometric radiographs of 495 patients were analyzed, collecting the horizontal and vertical dimensions of the hyoid bone, as well as the distance from the hyoid to the mandible. The following algorithms were trained: Logistic Regression, Gradient Boosting Classifier, K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Multilayer Perceptron Classifier (MLP), Decision Tree, AdaBoost Classifier, and Random Forest Classifier. A 5-fold cross-validation approach was used to validate each model. Model evaluation metrics included areas under the curve (AUC), accuracy, recall, precision, F1 score, and ROC curves. The horizontal dimension of the hyoid bone demonstrated the highest predictive power across all evaluated models. The AUC values of the different trained models ranged from 0.81 to 0.86 on test data and from 0.78 to 0.84 in cross-validation, with the random forest classifier achieving the highest accuracy rates. The supervised machine learning model showed good predictive accuracy, indicating the model's potential for sex determination in forensic and anthropological contexts. These findings suggest that the application of artificial intelligence methods can enhance the accuracy of sex estimation, contributing to significant advancements in the field.