Mahmut KURTBOĞAN, Gülçin RAY, Abdullah RAY
Experimental Biomedical Research - 2026;9(4):348-366
Aim: To assess the rate of sexual dimorphism in a healthy, geriatric Turkish population through the evaluation of selected mandibular variables on computed tomography (CT) scans where the cranium is displayed in three dimensions via machine learning algorithms ( MLA). Methods: The study sample consisted of 120 females and 120 males aged 65 or above. Fourteen linear mandibular variables were measured on three -dimensional CT images of the cranium, including right and left maximum ramus height (RMXRH, LMXRH) and bi -gonial breadth ( BGP). The machine learning algorithms utilised in the study are as follows; Random Forest (RandForest), Logistic Regression (LogReg), Linear Discriminant Analysis (LDA), Gradient Boosting (GradBoost), Support Vector Machine with Radial Basis Function Kerne l (SVM (RBF)), Support Vector Machine with Linear Kernel (SVM (Linear)), and K -Nearest Neighbors (KNN), Decision Tree (DecTree), Naive Bayes, and AdaBoost. Results: Statistically significant differences between sexes were found for all variables. The highest accuracy rate for MLA was obtained via Random Forest algorithm (Acc = 0.938). The LogReg, LDA, GradBoost, SVM (RBF), SVM (Linear), and KNN algorithms achieved the same accuracy rate (Acc = 0.917). The highest AUC values were observed in the LogReg and LDA algorithms based on the calculated ROC curve, with both models achieving an identical AUC score (AUC = 0.979). It was found that the most significant variable in sex prediction was RMXRH (importance score = 0.1686), followed by LMXRH (0.1234) and BGP (0.0908). Conclusion: It was found that the selected linear mandibular variables present high accuracy rates in determining the sex of geriatric Turkish population via MLA.
